mirror of
https://github.com/okieselbach/AutopilotMonitor.git
synced 2026-10-25 21:19:38 +00:00
Compare commits
7 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 48e9702fd2 | |||
| b560cb13b6 | |||
| b0f866dc7d | |||
| f3460ea828 | |||
| d88d4a143b | |||
| 913dfa13fe | |||
| 51345718c6 |
+1
-1
Submodule internal updated: fd299a3c0d...4b5afc55c1
+336
@@ -0,0 +1,336 @@
|
||||
using System;
|
||||
using System.IO;
|
||||
using System.Linq;
|
||||
using System.Text;
|
||||
using System.Threading.Tasks;
|
||||
using AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime;
|
||||
using Xunit;
|
||||
|
||||
namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
{
|
||||
/// <summary>
|
||||
/// B-96m: IME re-runs a failed platform script on later check-ins, so AgentExecutor.log can hold several runs of one
|
||||
/// policy before the tracker has read IME's result of the first — always on the agent's first pass (the agent starts
|
||||
/// mid-enrollment and reads all of AgentExecutor.log before IntuneManagementExtension.log), after a restart, and
|
||||
/// whenever a result line surfaces late. With one slot per policy the second executor start reused the first run's
|
||||
/// slot and its end block overwrote the first one: the first IME result went out with the LAST run's exit code and
|
||||
/// output (session eff1b413: the bootstrap script "Failed (exit 0)" with the stdout of its successful retry), every
|
||||
/// later result without an end block. These tests pin one run per executor start, each IME result paired with the
|
||||
/// run of its policy that started last before it, each run carrying its own end block.
|
||||
/// </summary>
|
||||
public sealed partial class ImeLogTrackerOverwriteRewindTests
|
||||
{
|
||||
private const string RetryPolicyId = "4c2d8e1a-7b3f-4a9e-8d21-6f0b5c3e9a74";
|
||||
private const string RetryUser = "00000000-0000-0000-0000-000000000000";
|
||||
|
||||
private static string ImeEntryAt(string time, string message)
|
||||
=> $"<![LOG[{message}]LOG]!><time=\"{time}\" date=\"9-10-2026\" component=\"IntuneManagementExtension\" context=\"\" type=\"1\" thread=\"12\" file=\"\">\r\n";
|
||||
|
||||
/// <summary>One executor run's start block: banner, the platform argument line, paths, launch.</summary>
|
||||
private static string RunStart(string time, string pid, string policyId = RetryPolicyId, string userId = RetryUser)
|
||||
{
|
||||
var path = $@"{ImeDir}\Policies\Scripts\{userId}_{policyId}.ps1";
|
||||
var results = $@"{ImeDir}\Policies\Results\{userId}_{policyId}";
|
||||
return ExecutorBanner(time)
|
||||
+ EntryAt(time, $"Adding argument powershell with value {path} to the named argument list.")
|
||||
+ EntryAt(time, "Powershell option gets invoked")
|
||||
+ EntryAt(time, path)
|
||||
+ EntryAt(time, results + ".output")
|
||||
+ EntryAt(time, results + ".error")
|
||||
+ EntryAt(time, results + ".timeout")
|
||||
+ ExecutorLaunch(time, path, pid);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// One executor run's end block in AgentExecutor's order. "error from script =" logs the stderr buffer before the
|
||||
/// output readers finished (<paramref name="stderrSnapshot"/>); "write output done" logs the final texts, the
|
||||
/// error part being exactly the error file IME judges.
|
||||
/// </summary>
|
||||
private static string RunEnd(string time, int exitCode, string stdout, string stderr, string? stderrSnapshot = null)
|
||||
=> EntryAt(time, $"Powershell exit code is {exitCode}")
|
||||
+ EntryAt(time, $"length of out={stdout.Length}")
|
||||
+ EntryAt(time, $"length of error={stderr.Length}")
|
||||
+ EntryAt(time, "error from script =" + (stderrSnapshot ?? stderr))
|
||||
+ EntryAt(time, exitCode == 0 ? "Powershell script is successfully executed." : "Powershell script is failed to execute")
|
||||
+ EntryAt(time, $"write output done. output = {stdout}, error = {stderr}")
|
||||
+ EntryAt(time, "Revert Wow64FsRedirection")
|
||||
+ EntryAt(time, "Agent executor completed.");
|
||||
|
||||
private static string ImeRunStart(string time, string policyId = RetryPolicyId, string userId = RetryUser)
|
||||
=> ImeEntryAt(time, $@"Script file {ImeDir}\Policies\Scripts\{userId}_{policyId}.ps1 is generated.");
|
||||
|
||||
private static string ImeRunResult(string time, string result, string policyId = RetryPolicyId, string userId = RetryUser)
|
||||
=> ImeEntryAt(time, $"[PowerShell] User Id = {userId}, Policy id = {policyId}, policy result = {result}");
|
||||
|
||||
private static void AppendImeLog(Harness h, string text)
|
||||
=> File.AppendAllText(Path.Combine(Path.GetDirectoryName(h.LogPath)!, "IntuneManagementExtension.log"), text, new UTF8Encoding(false));
|
||||
|
||||
/// <summary>Result | exit | stdout | stderr — readable in an assertion failure; "-" no exit code, "~" never read.</summary>
|
||||
private static string Shape(ScriptExecutionState s)
|
||||
=> $"{s.Result}|{(s.ExitCode.HasValue ? s.ExitCode.Value.ToString() : "-")}|{s.Stdout ?? "~"}|{s.Stderr ?? "~"}";
|
||||
|
||||
/// <summary>
|
||||
/// Two poll passes as the agent runs them (read the files, then the flushes), the second past the result hold:
|
||||
/// a result missing its end block goes out by then.
|
||||
/// </summary>
|
||||
private static async Task PassAndSettle(Harness h)
|
||||
{
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
h.Now = h.Now.AddSeconds(6);
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task First_pass_pairs_each_ime_result_with_the_end_block_of_its_own_run()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
// IME's first run and two retries on later check-ins, all written before the agent started.
|
||||
h.Append(RunStart("12:00:10.0000000", "1001") + RunEnd("12:00:12.0000000", 1, "", "Run A failed")
|
||||
+ RunStart("12:10:10.0000000", "1002") + RunEnd("12:10:12.0000000", 1, "Run B output", "")
|
||||
+ RunStart("12:20:10.0000000", "1003") + RunEnd("12:20:12.0000000", 0, "Run C output", ""));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed")
|
||||
+ ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:12.1000000", "Success")
|
||||
+ ImeRunStart("12:20:08.0000000") + ImeRunResult("12:20:12.1000000", "Success"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|1||Run A failed", "Success|1|Run B output|", "Success|0|Run C output|" },
|
||||
h.Completed.Select(Shape).ToArray());
|
||||
Assert.All(h.Completed, s => Assert.Equal("ime_policy_result", s.ResultSource));
|
||||
Assert.True(h.Completed[0].StartedAtUtc < h.Completed[1].StartedAtUtc && h.Completed[1].StartedAtUtc < h.Completed[2].StartedAtUtc);
|
||||
Assert.Equal(3, h.Completed.Select(s => s.RunId).Distinct().Count());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Retry_of_the_bootstrap_script_never_lends_its_success_output_to_the_failed_first_attempt()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "2001") + RunEnd("12:00:20.0000000", 1, "attempt 1: download failed", "Invoke-WebRequest : The remote name could not be resolved")
|
||||
+ RunStart("12:10:10.0000000", "2002") + RunEnd("12:10:20.0000000", 0, "===== Bootstrap Completed Successfully =====", ""));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:20.1000000", "Failed")
|
||||
+ ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:20.1000000", "Success"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[]
|
||||
{
|
||||
"Failed|1|attempt 1: download failed|Invoke-WebRequest : The remote name could not be resolved",
|
||||
"Success|0|===== Bootstrap Completed Successfully =====|",
|
||||
}, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_killed_at_the_timeout_reports_failed_without_the_next_runs_end_block()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
// IME killed the first executor after 1800 s: its start block is all AgentExecutor.log holds of it.
|
||||
h.Append(RunStart("12:00:10.0000000", "3001")
|
||||
+ RunStart("12:40:10.0000000", "3002") + RunEnd("12:40:12.0000000", 0, "Run B output", ""));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:30:10.5000000", "Failed")
|
||||
+ ImeRunStart("12:40:08.0000000") + ImeRunResult("12:40:12.1000000", "Success"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|-|~|~", "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_whose_result_never_surfaced_falls_back_on_imes_rule_before_the_next_run()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "4001") + RunEnd("12:00:12.0000000", 0, "Run A output", "Run A wrote an error")
|
||||
+ RunStart("12:10:10.0000000", "4002") + RunEnd("12:10:12.0000000", 0, "Run B output", ""));
|
||||
// Run A's result line was lost; run B's is there.
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:12.1000000", "Success"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|0|Run A output|Run A wrote an error", "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
Assert.Equal(new[] { "agentexecutor_fallback", "ime_policy_result" }, h.Completed.Select(s => s.ResultSource).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_in_flight_survives_the_emit_of_the_run_before_it()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "5001") + RunEnd("12:00:12.0000000", 0, "Run A output", "")
|
||||
+ RunStart("12:10:10.0000000", "5002"));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Success") + ImeRunStart("12:10:08.0000000"));
|
||||
await PassAndSettle(h);
|
||||
h.Now = h.Now.AddSeconds(30);
|
||||
await h.Pass();
|
||||
|
||||
// Run B ends minutes later.
|
||||
h.Append(RunEnd("12:10:40.0000000", 0, "Run B output", ""));
|
||||
AppendImeLog(h, ImeRunResult("12:10:40.1000000", "Success"));
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Success|0|Run A output|", "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Shutdown_between_the_two_logs_flushes_each_run_with_its_own_end_block()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "6001") + RunEnd("12:00:12.0000000", 1, "", "Run A failed")
|
||||
+ RunStart("12:10:10.0000000", "6002") + RunEnd("12:10:12.0000000", 0, "Run B output", ""));
|
||||
// The pass is cancelled after AgentExecutor.log; the shutdown flush follows.
|
||||
await h.Pass();
|
||||
h.Tracker.FlushPendingPlatformScriptResults(h.Now, force: true);
|
||||
|
||||
h.Restart();
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed")
|
||||
+ ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:12.1000000", "Success"));
|
||||
await PassAndSettle(h);
|
||||
|
||||
// Each run went out with its own block and IME's verdict on it; the results read after the restart are
|
||||
// the same runs and change nothing.
|
||||
Assert.Equal(new[] { "Failed|1||Run A failed", "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
Assert.All(h.Completed, s => Assert.Equal("agentexecutor_fallback", s.ResultSource));
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task End_block_line_recovered_after_the_next_run_started_stays_with_its_own_run()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "7001"));
|
||||
var exitLineAt = h.Length;
|
||||
h.Append(RunEnd("12:00:12.0000000", 1, "", "Run A failed")
|
||||
+ RunStart("12:10:10.0000000", "7002") + RunEnd("12:10:12.0000000", 0, "Run B output", ""));
|
||||
await h.Pass();
|
||||
|
||||
// A stale writer rewrites run A's exit line in place: the ledger check re-reads it behind run B's start.
|
||||
h.Overwrite(exitLineAt, EntryAt("12:00:12.0000000", "Powershell exit code is 7"));
|
||||
h.Tracker.RequestOverwriteCheckForTest();
|
||||
h.Now = h.Now.AddSeconds(2);
|
||||
await h.Pass();
|
||||
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed")
|
||||
+ ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:12.1000000", "Success"));
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|7||Run A failed", "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_without_result_or_end_block_is_dropped_once_a_later_run_has_its_result()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "9001")
|
||||
+ RunStart("12:10:10.0000000", "9002") + RunEnd("12:10:12.0000000", 0, "Run B output", ""));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunStart("12:10:08.0000000") + ImeRunResult("12:10:12.1000000", "Success"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Success|0|Run B output|" }, h.Completed.Select(Shape).ToArray());
|
||||
Assert.Equal(0, h.Tracker.ParkedPlatformScriptCountForTest);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public void Reading_the_same_executor_start_line_again_never_parks_its_run()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
var line = $@"Adding argument powershell with value {ImeDir}\Policies\Scripts\{RetryUser}_{RetryPolicyId}.ps1 to the named argument list.";
|
||||
var at = T0.AddMinutes(-30);
|
||||
|
||||
h.Tracker.ProcessLogMessageForTest(line, at, "AgentExecutor.log");
|
||||
h.Tracker.ProcessLogMessageForTest(line, at, "AgentExecutor.log");
|
||||
Assert.Equal(0, h.Tracker.ParkedPlatformScriptCountForTest);
|
||||
|
||||
h.Tracker.ProcessLogMessageForTest(line, at.AddMinutes(10), "AgentExecutor.log");
|
||||
Assert.Equal(1, h.Tracker.ParkedPlatformScriptCountForTest);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_without_end_block_keeps_waiting_for_its_result_across_a_shutdown()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "9101") + RunStart("12:40:10.0000000", "9102"));
|
||||
await h.Pass();
|
||||
h.Tracker.FlushPendingPlatformScriptResults(h.Now, force: true);
|
||||
Assert.Empty(h.Completed);
|
||||
|
||||
h.Restart();
|
||||
Assert.Equal(1, h.Tracker.ParkedPlatformScriptCountForTest);
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:30:10.5000000", "Failed"));
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|-|~|~" }, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Earlier_run_emitted_while_a_later_one_runs_never_takes_the_later_runs_register_entry()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "9201") + RunEnd("12:00:12.0000000", 1, "", "Run A failed")
|
||||
+ RunStart("12:10:10.0000000", "9202"));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed") + ImeRunStart("12:10:08.0000000"));
|
||||
await PassAndSettle(h);
|
||||
Assert.Single(h.Completed);
|
||||
|
||||
// A save of this key is the latest run's, and that one is still running.
|
||||
h.Tracker.ScriptRuns.NotePassCompleted(h.Now);
|
||||
Assert.Equal(PlatformRunClaim.Wait, h.Tracker.ScriptRuns.TryClaim(RetryUser, RetryPolicyId, h.Now.AddSeconds(-10), out _));
|
||||
|
||||
h.Append(RunEnd("12:10:40.0000000", 0, "Run B output", ""));
|
||||
AppendImeLog(h, ImeRunResult("12:10:40.1000000", "Success"));
|
||||
await PassAndSettle(h);
|
||||
h.Tracker.ScriptRuns.NotePassCompleted(h.Now);
|
||||
|
||||
Assert.Equal(PlatformRunClaim.Claimed, h.Tracker.ScriptRuns.TryClaim(RetryUser, RetryPolicyId, h.Now.AddSeconds(-10), out var run));
|
||||
Assert.Equal(h.Completed[1].RunId, run!.RunId);
|
||||
Assert.Equal("Run B output", run.Stdout);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Run_of_another_user_keeps_its_own_register_entry_when_it_waited_for_its_result()
|
||||
{
|
||||
const string otherUser = "2b7e5c91-3d4a-4f80-b6e2-9a1c7d3f5e08";
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "9301") + RunEnd("12:00:12.0000000", 1, "", "Run A failed")
|
||||
+ RunStart("12:10:10.0000000", "9302", userId: otherUser));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed")
|
||||
+ ImeRunStart("12:10:08.0000000", userId: otherUser));
|
||||
await PassAndSettle(h);
|
||||
h.Tracker.ScriptRuns.NotePassCompleted(h.Now);
|
||||
|
||||
Assert.Equal(PlatformRunClaim.Claimed, h.Tracker.ScriptRuns.TryClaim(RetryUser, RetryPolicyId, h.Now.AddSeconds(-10), out var run));
|
||||
Assert.Equal(h.Completed[0].RunId, run!.RunId);
|
||||
Assert.Equal(PlatformRunClaim.Wait, h.Tracker.ScriptRuns.TryClaim(otherUser, RetryPolicyId, h.Now.AddSeconds(-10), out _));
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Start_line_read_again_stays_a_late_line_after_an_earlier_run_of_another_user_went_out()
|
||||
{
|
||||
const string otherUser = "2b7e5c91-3d4a-4f80-b6e2-9a1c7d3f5e08";
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
h.Append(RunStart("12:00:10.0000000", "9401") + RunEnd("12:00:12.0000000", 0, "Run A output", "")
|
||||
+ RunStart("12:10:10.0000000", "9402", userId: otherUser) + RunEnd("12:10:12.0000000", 0, "Run B output", ""));
|
||||
// Run A's result surfaces after run B's (its line was hidden by a concurrent writer).
|
||||
AppendImeLog(h, ImeRunResult("12:10:12.1000000", "Success", userId: otherUser) + ImeRunResult("12:00:12.1000000", "Success"));
|
||||
await PassAndSettle(h);
|
||||
Assert.Equal(new[] { "Success|0|Run B output|", "Success|0|Run A output|" }, h.Completed.Select(Shape).ToArray());
|
||||
|
||||
// Run B's executor start line read again (a rewind): a late line of the emitted run B, never a new run.
|
||||
var started = 0;
|
||||
h.Tracker.OnScriptStarted = _ => started++;
|
||||
h.Tracker.ProcessLogMessageForTest(
|
||||
$@"Adding argument powershell with value {ImeDir}\Policies\Scripts\{otherUser}_{RetryPolicyId}.ps1 to the named argument list.",
|
||||
h.Completed[0].StartedAtUtc, "AgentExecutor.log");
|
||||
Assert.Equal(0, started);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Final_stderr_of_the_end_block_replaces_the_early_snapshot_even_without_stdout()
|
||||
{
|
||||
using var h = new Harness("AgentExecutor.log", ScriptPatterns());
|
||||
// The snapshot was taken before the error reader finished; IME judged the full error file.
|
||||
h.Append(RunStart("12:00:10.0000000", "8001") + RunEnd("12:00:12.0000000", 0, "", "Remove-AppxPackage : Cannot process argument", stderrSnapshot: ""));
|
||||
AppendImeLog(h, ImeRunStart("12:00:08.0000000") + ImeRunResult("12:00:12.1000000", "Failed"));
|
||||
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(new[] { "Failed|0||Remove-AppxPackage : Cannot process argument" }, h.Completed.Select(Shape).ToArray());
|
||||
}
|
||||
}
|
||||
}
|
||||
+139
@@ -0,0 +1,139 @@
|
||||
using System;
|
||||
using System.IO;
|
||||
using System.Linq;
|
||||
using System.Text;
|
||||
using System.Threading.Tasks;
|
||||
using AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime;
|
||||
using AutopilotMonitor.Agent.V2.Core.SignalAdapters;
|
||||
using Xunit;
|
||||
|
||||
namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
{
|
||||
/// <summary>
|
||||
/// Sessions d29e27e3 and 4377911b: a concurrent IME writer overwrote the result line of a platform script, and the
|
||||
/// overwrite check recovered it one pass later. Recovered lines were never fresh, so they resolved in the agent's own
|
||||
/// cached zone (+60) instead of anchoring themselves (+120): the end stamp landed an hour ahead of the clock and the
|
||||
/// run duration was omitted. Bytes that read differently at a position than at the last look were written after that
|
||||
/// look, so when the replaced bytes were confirmed seconds ago the new line's age is bounded like an appended line's —
|
||||
/// these tests pin that it anchors itself then, and only then.
|
||||
/// </summary>
|
||||
public sealed partial class ImeLogTrackerOverwriteRewindTests
|
||||
{
|
||||
private static void AppendExecutorLog(Harness h, string text)
|
||||
=> File.AppendAllText(Path.Combine(Path.GetDirectoryName(h.LogPath)!, "AgentExecutor.log"), text, new UTF8Encoding(false));
|
||||
|
||||
/// <summary>An IME log entry with the byte length of <paramref name="sameLengthAs"/>: what a stale writer's line replaces exactly.</summary>
|
||||
private static string FillerLike(string time, string sameLengthAs)
|
||||
{
|
||||
var bare = ImeEntryAt(time, "[Win32App] checking ");
|
||||
Assert.True(sameLengthAs.Length > bare.Length);
|
||||
return ImeEntryAt(time, "[Win32App] checking " + new string('x', sameLengthAs.Length - bare.Length));
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// A platform run whose IME writer lives in +120 (local = UTC + 2 h; the harness clock is 19:30Z): both logs are
|
||||
/// observed empty first, so what follows is read fresh. Returns the IME log position where the filler line sits.
|
||||
/// </summary>
|
||||
private static async Task<long> StartRunAndReadFiller(Harness h, string fillerTime, string resultEntry)
|
||||
{
|
||||
h.Append(string.Empty);
|
||||
AppendExecutorLog(h, string.Empty);
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
|
||||
AppendExecutorLog(h, RunStart("21:29:50.0000000", "1001"));
|
||||
h.Append(ImeRunStart("21:29:48.0000000"));
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
|
||||
// Another IME writer appends a line; the tracker reads it.
|
||||
var at = h.Length;
|
||||
h.Append(FillerLike(fillerTime, resultEntry));
|
||||
h.Now = h.Now.AddSeconds(1);
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
return at;
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Result_line_recovered_seconds_after_its_bytes_were_read_anchors_itself_like_a_fresh_line()
|
||||
{
|
||||
using var h = new Harness("IntuneManagementExtension.log", ScriptPatterns());
|
||||
var result = ImeRunResult("21:30:01.1000000", "Success");
|
||||
var at = await StartRunAndReadFiller(h, "21:30:00.5000000", result);
|
||||
|
||||
// The script ends: its end block, then the stale IME service writes the result over the filler.
|
||||
AppendExecutorLog(h, RunEnd("21:30:01.0000000", 0, "done", ""));
|
||||
h.Overwrite(at, result);
|
||||
h.Now = h.Now.AddSeconds(1);
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.Equal(1, h.Health.OverwriteRewinds);
|
||||
var script = Assert.Single(h.Completed);
|
||||
Assert.Equal(CmTraceOffsetOrigin.LineAnchored, script.ResultProvenance!.Origin);
|
||||
Assert.Equal(120, script.ResultProvenance.OffsetMinutes);
|
||||
Assert.Equal(ImeLogTrackerAdapter.DurationPairVerdict.Verified, ImeLogTrackerAdapter.JudgeDurationPair(
|
||||
script.StartedAtProvenance, script.ResultProvenance, false, script.ResultObservedAtUtc!.Value - script.StartedAtUtc!.Value, out _));
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Verification_passes_keep_the_replaced_bytes_current_for_a_script_that_runs_longer()
|
||||
{
|
||||
using var h = new Harness("IntuneManagementExtension.log", ScriptPatterns());
|
||||
var result = ImeRunResult("21:30:41.1000000", "Success");
|
||||
var at = await StartRunAndReadFiller(h, "21:30:00.5000000", result);
|
||||
|
||||
// The script runs 40 s; the cadence check verifies the ledger every 10 s while the run is in flight.
|
||||
for (var i = 0; i < 4; i++)
|
||||
{
|
||||
h.Now = h.Now.AddSeconds(10);
|
||||
await h.Tracker.RunPollPassAsync(default);
|
||||
}
|
||||
|
||||
AppendExecutorLog(h, RunEnd("21:30:41.0000000", 0, "done", ""));
|
||||
h.Overwrite(at, result);
|
||||
h.Now = h.Now.AddSeconds(1);
|
||||
await PassAndSettle(h);
|
||||
|
||||
var script = Assert.Single(h.Completed);
|
||||
Assert.Equal(CmTraceOffsetOrigin.LineAnchored, script.ResultProvenance!.Origin);
|
||||
Assert.Equal(120, script.ResultProvenance.OffsetMinutes);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public async Task Overwrite_does_not_move_the_files_measured_writer_offset()
|
||||
{
|
||||
using var h = new Harness("IntuneManagementExtension.log", ScriptPatterns());
|
||||
var result = ImeRunResult("21:30:01.1000000", "Success");
|
||||
var at = await StartRunAndReadFiller(h, "21:30:00.5000000", result);
|
||||
Assert.True(h.Tracker.OffsetCalibrator.TryGetOffset("IntuneManagementExtension.log", out var before));
|
||||
|
||||
AppendExecutorLog(h, RunEnd("21:30:01.0000000", 0, "done", ""));
|
||||
h.Overwrite(at, result);
|
||||
h.Now = h.Now.AddSeconds(1);
|
||||
await PassAndSettle(h);
|
||||
|
||||
Assert.True(h.Tracker.OffsetCalibrator.TryGetOffset("IntuneManagementExtension.log", out var after));
|
||||
Assert.Equal(before, after);
|
||||
}
|
||||
|
||||
[Theory]
|
||||
// Replaced bytes confirmed 1 s / 30 s ago: written since, age bounded like an appended line's.
|
||||
[InlineData(true, true, 1, true)]
|
||||
[InlineData(true, true, 30, true)]
|
||||
// Confirmed longer ago: the new bytes could be minutes old — their age may round onto the 15-minute grid.
|
||||
[InlineData(true, true, 31, false)]
|
||||
// The clock stepped back since the confirmation: the age is unknown.
|
||||
[InlineData(true, true, -1, false)]
|
||||
// The divergence lies in bytes the ledger never held (dropped by a cap): nothing dates them.
|
||||
[InlineData(true, false, 1, false)]
|
||||
// The pass itself is not fresh — the first pass after a restart never anchors.
|
||||
[InlineData(false, true, 1, false)]
|
||||
// Nothing was replaced.
|
||||
[InlineData(true, true, null, false)]
|
||||
public void Rewound_entry_is_fresh_only_when_the_bytes_it_replaced_were_confirmed_within_the_fresh_window(
|
||||
bool passLinesAreFresh, bool entryAtRewindPointReplaced, int? secondsSinceConfirmed, bool expected)
|
||||
{
|
||||
var now = new DateTime(2026, 10, 4, 19, 2, 0, DateTimeKind.Utc);
|
||||
DateTime? confirmed = secondsSinceConfirmed.HasValue ? now.AddSeconds(-secondsSinceConfirmed.Value) : (DateTime?)null;
|
||||
Assert.Equal(expected, ImeLogTracker.RewoundEntryIsFresh(passLinesAreFresh, entryAtRewindPointReplaced, confirmed, now));
|
||||
}
|
||||
}
|
||||
}
|
||||
+4
@@ -502,6 +502,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
public async Task Ime_result_line_hidden_by_an_executor_start_line_is_recovered_from_the_ime_log()
|
||||
{
|
||||
using var h = new Harness("IntuneManagementExtension.log", ScriptPatterns());
|
||||
// Seen empty first: the start lines are fresh and anchor themselves like the recovered result line will,
|
||||
// so start and result sit on one timeline whatever the host's own zone is.
|
||||
h.Append(string.Empty);
|
||||
await h.Pass();
|
||||
h.Append(Entry(GeneratedLine, "17") + Entry("Launch powershell executor in user session", "17") + Entry("process id = 10608", "17"));
|
||||
await h.Pass();
|
||||
var servicePosition = h.Length;
|
||||
|
||||
+26
-1
@@ -83,8 +83,33 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
Assert.Single(emitted);
|
||||
}
|
||||
|
||||
[Theory]
|
||||
// IME's verdict is Success exactly when the error file is empty, whatever the exit code. Judged by the exit
|
||||
// code, session 67ae751c showed exit 0 with the script's own stderr as Success, and 46a41a0a an exit 1 without
|
||||
// stderr as Failed, a run IME reported as Success one cycle earlier.
|
||||
[InlineData(0, "Invoke-WebRequest : The remote name could not be resolved", "Failed")]
|
||||
[InlineData(1, "", "Success")]
|
||||
[InlineData(1, " \n", "Success")]
|
||||
[InlineData(0, "", "Success")]
|
||||
[InlineData(1, null, "Failed")]
|
||||
[InlineData(0, null, "Success")]
|
||||
public void Fallback_judges_like_ime_by_the_stderr_it_read(int exitCode, string? stderr, string expected)
|
||||
{
|
||||
using var tmp = new TempDirectory();
|
||||
var tracker = BuildTracker(tmp, out var emitted);
|
||||
var observedAt = new DateTime(2026, 6, 19, 12, 59, 4, DateTimeKind.Utc);
|
||||
|
||||
tracker.SeedPendingPlatformScriptForTesting("0d2294bf", exitCode: exitCode, exitObservedAtUtc: observedAt, stderr: stderr);
|
||||
tracker.FlushPendingPlatformScriptResults(observedAt.AddSeconds(20));
|
||||
|
||||
var script = Assert.Single(emitted);
|
||||
Assert.Equal(expected, script.Result);
|
||||
Assert.Equal(exitCode, script.ExitCode);
|
||||
Assert.Equal("agentexecutor_fallback", script.ResultSource);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public void Nonzero_exit_code_emits_failed_result()
|
||||
public void Nonzero_exit_code_without_a_stderr_line_emits_failed_result()
|
||||
{
|
||||
using var tmp = new TempDirectory();
|
||||
var tracker = BuildTracker(tmp, out var emitted);
|
||||
|
||||
+27
@@ -108,6 +108,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
|
||||
var tracker = BuildTracker(tmp, out var emitted);
|
||||
tracker.LoadStateForTest(); // must not throw
|
||||
Assert.Equal(0, tracker.ParkedPlatformScriptCountForTest);
|
||||
|
||||
// Degrades to pre-fix behavior: no markers restored, normal emission works (the
|
||||
// result waits for its end block until the shutdown flush).
|
||||
@@ -116,6 +117,32 @@ namespace AutopilotMonitor.Agent.V2.Core.Tests.Monitoring.Ime
|
||||
Assert.Single(emitted);
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public void Pending_slot_from_a_state_file_without_executor_start_loads_and_completes()
|
||||
{
|
||||
using var tmp = new TempDirectory();
|
||||
var started = new DateTime(2026, 10, 4, 9, 0, 0, DateTimeKind.Utc);
|
||||
|
||||
// A pending slot written by an agent before runs carried their executor start and run id.
|
||||
var persistence = new ImeTrackerStatePersistence(tmp.Path, new AgentLogger(tmp.Path, AgentLogLevel.Info));
|
||||
persistence.Save(new ImeTrackerStateData
|
||||
{
|
||||
PendingPlatformScripts = new List<ScriptExecutionState>
|
||||
{
|
||||
new ScriptExecutionState { PolicyId = "policyF", ScriptType = "platform", StartedAtUtc = started, ExitCode = 0, ExitObservedAtUtc = started.AddSeconds(5), Stdout = "done", Stderr = "" },
|
||||
},
|
||||
});
|
||||
|
||||
var tracker = BuildTracker(tmp, out var emitted);
|
||||
tracker.LoadStateForTest();
|
||||
tracker.CompletePlatformScriptFromImeResultForTesting("policyF", "Success", started.AddSeconds(6));
|
||||
|
||||
var script = Assert.Single(emitted);
|
||||
Assert.Equal("Success", script.Result);
|
||||
Assert.Equal(0, script.ExitCode);
|
||||
Assert.False(string.IsNullOrEmpty(script.RunId));
|
||||
}
|
||||
|
||||
[Fact]
|
||||
public void Stale_result_older_than_current_run_start_is_dropped()
|
||||
{
|
||||
|
||||
+254
-57
@@ -306,6 +306,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
{
|
||||
var currentStartTs = LastMatchedLogTimestamp ?? UtcNowProvider();
|
||||
var currentStartProvenance = CaptureLastMatchedProvenance();
|
||||
var fromExecutor = string.Equals(source, "agentexecutor", StringComparison.OrdinalIgnoreCase);
|
||||
|
||||
// A start line older than the result this policy already emitted is the late start
|
||||
// line of THAT run: its AgentExecutor block surfaced after the IME result — the two
|
||||
@@ -318,7 +319,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
// never tie; a low-resolution clock in a test may).
|
||||
if (_platformScriptResultEmitted.TryGetValue(id, out var emittedResultTs) && currentStartTs < emittedResultTs)
|
||||
{
|
||||
RecordInvocationMarker(id, isClose: false);
|
||||
RecordInvocationMarker(id, isClose: false, EmittedRunId);
|
||||
NoteInvocationActivity();
|
||||
_logger.Debug($"ImeLogTracker: platform script start line for {id} (source: {source ?? "ime"}) predates its emitted result — late line of that run, not a new one");
|
||||
return;
|
||||
@@ -326,7 +327,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
|
||||
if (_pendingPlatformScripts.TryGetValue(id, out var pendingSlot))
|
||||
{
|
||||
if (pendingSlot.StartedAtUtc.HasValue && currentStartTs - pendingSlot.StartedAtUtc.Value > TimeSpan.FromHours(24))
|
||||
if (pendingSlot.StartedAtUtc.HasValue && currentStartTs - pendingSlot.StartedAtUtc.Value > StalePlatformRunAge)
|
||||
{
|
||||
// Stale-slot hardening (session eaf3d8c4): a replayed start line from a previous
|
||||
// enrollment whose matching result never appeared (script was mid-run when that
|
||||
@@ -343,6 +344,15 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
// Emit what the held run has; its end block did not surface in time.
|
||||
EmitPlatformScriptResult(pendingSlot, "the next run of this policy started");
|
||||
}
|
||||
else if (fromExecutor && LastMatchedLogTimestamp.HasValue && pendingSlot.Result == null
|
||||
&& pendingSlot.ExecutorStartedAtUtc.HasValue && pendingSlot.ExecutorStartedAtUtc.Value < currentStartTs)
|
||||
{
|
||||
// Every executor process writes its start line once: a later one is the next run
|
||||
// (IME re-runs a failed script on a later check-in), whose end block must not
|
||||
// overwrite this one's. This run waits for its IME result — on the first pass
|
||||
// AgentExecutor.log is read in full before the IME log (session eff1b413).
|
||||
ParkPlatformRun(pendingSlot);
|
||||
}
|
||||
else if (!pendingSlot.StartedAtUtc.HasValue)
|
||||
{
|
||||
// The slot was opened by its result while the start block was still hidden;
|
||||
@@ -351,9 +361,17 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
pendingSlot.StartedAtProvenance = currentStartProvenance;
|
||||
_stateDirty = true;
|
||||
}
|
||||
if (pendingSlot.UserId == null && userId != null) pendingSlot.UserId = userId;
|
||||
}
|
||||
if (!_pendingPlatformScripts.ContainsKey(id))
|
||||
if (_pendingPlatformScripts.TryGetValue(id, out var slot))
|
||||
{
|
||||
if (slot.UserId == null && userId != null) slot.UserId = userId;
|
||||
if (fromExecutor && !slot.ExecutorStartedAtUtc.HasValue)
|
||||
{
|
||||
slot.ExecutorStartedAtUtc = currentStartTs;
|
||||
_stateDirty = true;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// A fresh start for this policy (no pending entry) begins a NEW execution.
|
||||
// Clear any emitted-marker from a prior run so IME re-evaluations / retries of
|
||||
@@ -361,18 +379,21 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
// deduped away. A start line of the emitted run itself never reaches this point
|
||||
// (the guard above), so this never clears the current run's own marker.
|
||||
_platformScriptResultEmitted.Remove(id);
|
||||
_pendingPlatformScripts[id] = new ScriptExecutionState
|
||||
slot = new ScriptExecutionState
|
||||
{
|
||||
PolicyId = id,
|
||||
ScriptType = "platform",
|
||||
UserId = userId,
|
||||
RunId = NewRunId(),
|
||||
// Prefer the source CMTrace timestamp so a script that started before the
|
||||
// agent launched (replayed log content) is dated to its real start, not now.
|
||||
// Set only at slot creation: the start line fires twice (agentexecutor + ime
|
||||
// source) but the earliest wins.
|
||||
StartedAtUtc = currentStartTs,
|
||||
StartedAtProvenance = currentStartProvenance,
|
||||
ExecutorStartedAtUtc = fromExecutor ? currentStartTs : (DateTime?)null,
|
||||
};
|
||||
_pendingPlatformScripts[id] = slot;
|
||||
ScriptRuns.NoteStarted(userId, id, UtcNowProvider());
|
||||
// Live "running" indicator — same signal health scripts emit. Gated on slot
|
||||
// creation so the duplicate start line (agentexecutor + ime source) doesn't
|
||||
@@ -380,7 +401,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
try { OnScriptStarted?.Invoke(new ScriptStartedInfo { PolicyId = id, ScriptType = "platform" }); }
|
||||
catch (Exception ex) { _logger.Warning($"ImeLogTracker: OnScriptStarted handler threw: {ex.Message}"); }
|
||||
}
|
||||
RecordInvocationMarker(id, isClose: false);
|
||||
RecordInvocationMarker(id, isClose: false, LastMatchedLogTimestamp.HasValue ? RunIdOfStartLine(slot, fromExecutor, currentStartTs) : slot.RunId);
|
||||
NoteInvocationActivity();
|
||||
// Started lines fire twice per script (agentexecutor + ime source) and carry no
|
||||
// outcome — the matching `platform script completed` line below carries result+exit
|
||||
@@ -388,6 +409,34 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
_logger.Debug($"ImeLogTracker: platform script started: {id} (source: {source ?? "ime"})");
|
||||
}
|
||||
|
||||
/// <summary>How long after IME's "Script file … is generated" line its executor writes its start line (about 2 s).</summary>
|
||||
private static readonly TimeSpan ImeLaunchWindow = TimeSpan.FromMinutes(1);
|
||||
|
||||
/// <summary>
|
||||
/// The run a timestamped start line belongs to, for its invocation marker. An executor start line belongs to the
|
||||
/// slot unless it is older than the slot's executor start — re-read behind a later run — then to the parked run
|
||||
/// it started. IME's "Script file … is generated" precedes its executor: it belongs to the live run whose
|
||||
/// executor started within <see cref="ImeLaunchWindow"/> after it, else to a slot whose executor has not
|
||||
/// started yet; otherwise its run is not known (yet) and the lines behind it go nowhere.
|
||||
/// </summary>
|
||||
private string RunIdOfStartLine(ScriptExecutionState slot, bool fromExecutor, DateTime lineTs)
|
||||
{
|
||||
if (fromExecutor)
|
||||
{
|
||||
if (!slot.ExecutorStartedAtUtc.HasValue || slot.ExecutorStartedAtUtc.Value <= lineTs) return slot.RunId;
|
||||
var parked = _parkedPlatformScripts.FirstOrDefault(r =>
|
||||
string.Equals(r.PolicyId, slot.PolicyId, StringComparison.OrdinalIgnoreCase) && r.ExecutorStartedAtUtc == lineTs);
|
||||
return parked?.RunId ?? EmittedRunId;
|
||||
}
|
||||
|
||||
var launched = LivePlatformRuns(slot.PolicyId)
|
||||
.Where(r => r.ExecutorStartedAtUtc.HasValue && r.ExecutorStartedAtUtc.Value >= lineTs && r.ExecutorStartedAtUtc.Value - lineTs <= ImeLaunchWindow)
|
||||
.OrderBy(r => r.ExecutorStartedAtUtc.Value)
|
||||
.FirstOrDefault();
|
||||
if (launched != null) return launched.RunId;
|
||||
return slot.ExecutorStartedAtUtc.HasValue ? EmittedRunId : slot.RunId;
|
||||
}
|
||||
|
||||
// Routes line-by-line script-data handlers (context / exitCode / output) to the right
|
||||
// accumulator based on the pattern's scriptType parameter. Health-script lines feed the
|
||||
// single-slot _pendingHealthScript (the HS-NEW-RESULT JSON delivers the full cycle in one
|
||||
@@ -502,13 +551,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
|
||||
// PS-AGENT-OUTPUT captures both stdout and stderr in one pattern
|
||||
var output = match.Groups["output"]?.Value;
|
||||
var error = match.Groups["error"]?.Value;
|
||||
var error = match.Groups["error"];
|
||||
|
||||
if (!string.IsNullOrEmpty(output) && !string.IsNullOrEmpty(error))
|
||||
if (error != null && error.Success)
|
||||
{
|
||||
// Combined pattern (write output done. output = ..., error = ...)
|
||||
script.Stdout = TruncateOutput(output.Trim());
|
||||
script.Stderr = TruncateOutput(error.Trim());
|
||||
// Combined pattern (write output done. output = ..., error = ...): both texts as
|
||||
// AgentExecutor wrote them to the result files — the error part is the error file
|
||||
// IME judges. The "error from script" line before it is a snapshot taken before the
|
||||
// output readers finished, so this one replaces it, an empty stdout included.
|
||||
script.Stdout = TruncateOutput(output?.Trim());
|
||||
script.Stderr = TruncateOutput(error.Value.Trim());
|
||||
}
|
||||
else if (string.Equals(outputType, "stderr", StringComparison.OrdinalIgnoreCase))
|
||||
{
|
||||
@@ -556,49 +608,44 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// </summary>
|
||||
internal void CompletePlatformScriptFromImeResult(string policyId, string result, DateTime? resultLineTimestampUtc, string patternId, CmTraceLineProvenance resultProvenance, string userId = null)
|
||||
{
|
||||
// Dedup: the deadline-based fallback (FlushPendingPlatformScriptResults) may already have
|
||||
// emitted this script from its AgentExecutor exit code because IME's authoritative
|
||||
// PS-SCRIPT-RESULT line arrived late. The fallback carries the same exit code + stdout,
|
||||
// so re-emitting here would duplicate the timeline entry and inflate counts. Drop the
|
||||
// now-redundant pending entry and skip.
|
||||
if (_platformScriptResultEmitted.ContainsKey(policyId))
|
||||
// The result belongs to the run of this policy that started last before its line —
|
||||
// the run in flight, or an earlier one that waits for its result while a later run
|
||||
// already started (ParkPlatformRun).
|
||||
var script = FindRunForResult(policyId, userId, resultLineTimestampUtc);
|
||||
if (script == null)
|
||||
{
|
||||
_logger.Debug($"ImeLogTracker: platform script {policyId} PS-SCRIPT-RESULT arrived after fallback emit — skipping duplicate");
|
||||
_pendingPlatformScripts.Remove(policyId);
|
||||
return;
|
||||
}
|
||||
// Dedup: the deadline-based fallback (FlushPendingPlatformScriptResults) may already
|
||||
// have emitted this run from its AgentExecutor end block because IME's authoritative
|
||||
// PS-SCRIPT-RESULT line arrived late, or the same result line is read again.
|
||||
// Re-emitting would duplicate the timeline entry and inflate counts.
|
||||
if (_platformScriptResultEmitted.ContainsKey(policyId))
|
||||
{
|
||||
_logger.Debug($"ImeLogTracker: platform script {policyId} PS-SCRIPT-RESULT arrived after its run was emitted — skipping duplicate");
|
||||
return;
|
||||
}
|
||||
|
||||
// Merge with pending AgentExecutor data if available
|
||||
if (_pendingPlatformScripts.TryGetValue(policyId, out var script))
|
||||
{
|
||||
// Stale-result guard: on an intra-lifetime re-run of the same policy, run 1's
|
||||
// late PS-SCRIPT-RESULT can arrive after run 2's start line already created a
|
||||
// fresh slot (the start clears the emitted-marker). Merging would pair run 2's
|
||||
// StartedAtUtc/state with run 1's result and then drop run 2's genuine result at
|
||||
// the marker check above. A result line always postdates its own run's start
|
||||
// line in the source log, so a result older than the pending slot's start
|
||||
// belongs to a previous run — drop it and keep the live slot intact.
|
||||
if (script.StartedAtUtc.HasValue
|
||||
&& resultLineTimestampUtc.HasValue
|
||||
&& resultLineTimestampUtc.Value < script.StartedAtUtc.Value)
|
||||
// Stale-result guard: every run of this policy in flight started after the result
|
||||
// line — a result line always postdates its own run's start line in the source log,
|
||||
// so it belongs to a previous run. Keep the live runs intact.
|
||||
if (HasLivePlatformRun(policyId))
|
||||
{
|
||||
_logger.Debug($"ImeLogTracker: platform script {policyId} result predates the current run's start — stale result from a previous run, skipping");
|
||||
return;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
script = new ScriptExecutionState
|
||||
{
|
||||
PolicyId = policyId,
|
||||
ScriptType = "platform",
|
||||
UserId = userId,
|
||||
RunId = NewRunId(),
|
||||
};
|
||||
_pendingPlatformScripts[policyId] = script;
|
||||
ScriptRuns.NoteStarted(userId, policyId, UtcNowProvider());
|
||||
}
|
||||
|
||||
if (script.UserId == null && userId != null) script.UserId = userId;
|
||||
ResolveParkedRunsBefore(script);
|
||||
script.Result = result;
|
||||
script.ResultSource = "ime_policy_result";
|
||||
script.ResultObservedAtUtc = resultLineTimestampUtc ?? UtcNowProvider();
|
||||
@@ -627,21 +674,150 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
private void EmitPlatformScriptResult(ScriptExecutionState script, string note = null)
|
||||
{
|
||||
_logger.Info($"ImeLogTracker: platform script completed: {script.PolicyId}, result={script.Result}, exit={script.ExitCode}{(note != null ? $" ({note})" : string.Empty)}");
|
||||
if (string.IsNullOrEmpty(script.RunId)) script.RunId = Guid.NewGuid().ToString("D");
|
||||
if (string.IsNullOrEmpty(script.RunId)) script.RunId = NewRunId();
|
||||
EmitScriptEvent(script);
|
||||
ScriptRuns.NoteEmitted(script, UtcNowProvider());
|
||||
// The register keeps the latest run per (user, policy) for the registry reconciliation:
|
||||
// a run that waited for its result while a later one started must not take that later
|
||||
// run's entry.
|
||||
if (!HasLaterLivePlatformRun(script)) ScriptRuns.NoteEmitted(script, UtcNowProvider());
|
||||
// The run's source stamp, capped at the clock: a result line cannot be read before it
|
||||
// was written, so a stamp ahead of the clock is a wrong zone assumption on that line
|
||||
// (a rewound result resolved in the reader's zone, session 4377911b: +1 h). Left
|
||||
// uncapped, the marker made the genuine NEXT run's start line read as a late line of
|
||||
// this run and its result a duplicate.
|
||||
// this run and its result a duplicate. Only ever raised: a run emitted after a later
|
||||
// one must not move the reference back.
|
||||
var runStamp = script.ResultObservedAtUtc ?? script.ExitObservedAtUtc ?? UtcNowProvider();
|
||||
var nowUtc = UtcNowProvider();
|
||||
_platformScriptResultEmitted[script.PolicyId] = runStamp > nowUtc ? nowUtc : runStamp;
|
||||
_pendingPlatformScripts.Remove(script.PolicyId);
|
||||
var stamp = runStamp > nowUtc ? nowUtc : runStamp;
|
||||
if (!_platformScriptResultEmitted.TryGetValue(script.PolicyId, out var previousStamp) || stamp > previousStamp)
|
||||
_platformScriptResultEmitted[script.PolicyId] = stamp;
|
||||
if (_pendingPlatformScripts.TryGetValue(script.PolicyId, out var current) && ReferenceEquals(current, script))
|
||||
_pendingPlatformScripts.Remove(script.PolicyId);
|
||||
else
|
||||
_parkedPlatformScripts.Remove(script);
|
||||
_stateDirty = true;
|
||||
}
|
||||
|
||||
private static string NewRunId() => Guid.NewGuid().ToString("D");
|
||||
|
||||
/// <summary>A run older than this by source time is from a previous enrollment, not a run in flight.</summary>
|
||||
private static readonly TimeSpan StalePlatformRunAge = TimeSpan.FromHours(24);
|
||||
|
||||
private static bool SameUserOrUnknown(string a, string b)
|
||||
=> a == null || b == null || string.Equals(a, b, StringComparison.OrdinalIgnoreCase);
|
||||
|
||||
private static DateTime StartOf(ScriptExecutionState run) => run.StartedAtUtc ?? DateTime.MinValue;
|
||||
|
||||
/// <summary>The runs of a policy that are not emitted yet: the earlier ones waiting for their IME result, then the one in flight.</summary>
|
||||
private IEnumerable<ScriptExecutionState> LivePlatformRuns(string policyId)
|
||||
{
|
||||
foreach (var run in _parkedPlatformScripts)
|
||||
{
|
||||
if (string.Equals(run.PolicyId, policyId, StringComparison.OrdinalIgnoreCase)) yield return run;
|
||||
}
|
||||
if (_pendingPlatformScripts.TryGetValue(policyId, out var current)) yield return current;
|
||||
}
|
||||
|
||||
private bool HasLivePlatformRun(string policyId) => LivePlatformRuns(policyId).Any();
|
||||
|
||||
/// <summary>The live run a start marker names, or null once that run was emitted.</summary>
|
||||
private ScriptExecutionState FindLivePlatformRun(string policyId, string runId)
|
||||
=> LivePlatformRuns(policyId).FirstOrDefault(r => string.Equals(r.RunId, runId, StringComparison.Ordinal));
|
||||
|
||||
private bool HasLaterLivePlatformRun(ScriptExecutionState script)
|
||||
=> LivePlatformRuns(script.PolicyId).Any(r => !ReferenceEquals(r, script)
|
||||
&& SameUserOrUnknown(r.UserId, script.UserId) && StartOf(r) > StartOf(script));
|
||||
|
||||
/// <summary>
|
||||
/// The live run an IME result line belongs to: of the runs of its policy that started at or before the line,
|
||||
/// the latest — one of the same IME user when there is one.
|
||||
/// </summary>
|
||||
private ScriptExecutionState FindRunForResult(string policyId, string userId, DateTime? resultLineTimestampUtc)
|
||||
{
|
||||
ScriptExecutionState sameUser = null, anyUser = null;
|
||||
foreach (var run in LivePlatformRuns(policyId))
|
||||
{
|
||||
if (resultLineTimestampUtc.HasValue && run.StartedAtUtc.HasValue && run.StartedAtUtc.Value > resultLineTimestampUtc.Value) continue;
|
||||
if (anyUser == null || StartOf(run) > StartOf(anyUser)) anyUser = run;
|
||||
if (SameUserOrUnknown(run.UserId, userId) && (sameUser == null || StartOf(run) > StartOf(sameUser))) sameUser = run;
|
||||
}
|
||||
return sameUser ?? anyUser;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Moves the run in flight aside when the next executor run of its policy starts: it keeps its own end block and
|
||||
/// waits for its IME result (<see cref="FindRunForResult"/>), the exit-code fallback, or a later run's result.
|
||||
/// </summary>
|
||||
private void ParkPlatformRun(ScriptExecutionState run)
|
||||
{
|
||||
_pendingPlatformScripts.Remove(run.PolicyId);
|
||||
_parkedPlatformScripts.Add(run);
|
||||
_stateDirty = true;
|
||||
if (_parkedPlatformScripts.Count > MaxParkedPlatformScripts)
|
||||
{
|
||||
var oldest = _parkedPlatformScripts.OrderBy(StartOf).First();
|
||||
_logger.Warning($"ImeLogTracker: more than {MaxParkedPlatformScripts} platform runs wait for their IME result — resolving the oldest ({oldest.PolicyId}, started {oldest.StartedAtUtc:o}) now");
|
||||
ResolveParkedRun(oldest, "too many runs waiting for their IME result");
|
||||
}
|
||||
_logger.Debug($"ImeLogTracker: platform script {run.PolicyId}: the next executor run started before IME's result of run {run.RunId} was read — that run keeps its end block and waits for its result");
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Results arrive in run order: a parked run of the same IME user that started before the run a result belongs to
|
||||
/// gets none any more (its result line was lost) — it goes out now, ahead of the later run.
|
||||
/// </summary>
|
||||
private void ResolveParkedRunsBefore(ScriptExecutionState owner)
|
||||
{
|
||||
if (_parkedPlatformScripts.Count == 0 || !owner.StartedAtUtc.HasValue) return;
|
||||
var earlier = _parkedPlatformScripts
|
||||
.Where(r => !ReferenceEquals(r, owner)
|
||||
&& string.Equals(r.PolicyId, owner.PolicyId, StringComparison.OrdinalIgnoreCase)
|
||||
&& SameUserOrUnknown(r.UserId, owner.UserId)
|
||||
&& r.StartedAtUtc.HasValue && r.StartedAtUtc.Value < owner.StartedAtUtc.Value)
|
||||
.OrderBy(StartOf)
|
||||
.ToList();
|
||||
foreach (var run in earlier) ResolveParkedRun(run, "a later run of this policy got its IME result first");
|
||||
}
|
||||
|
||||
/// <summary>A parked run that will not get its IME result: out with what it has, or dropped when it has nothing.</summary>
|
||||
private void ResolveParkedRun(ScriptExecutionState run, string why)
|
||||
{
|
||||
if (run.Result != null)
|
||||
{
|
||||
EmitPlatformScriptResult(run, why);
|
||||
}
|
||||
else if (run.ExitCode.HasValue)
|
||||
{
|
||||
EmitFromExecutorEndBlock(run, why);
|
||||
}
|
||||
else
|
||||
{
|
||||
_parkedPlatformScripts.Remove(run);
|
||||
_stateDirty = true;
|
||||
_logger.Debug($"ImeLogTracker: platform script {run.PolicyId} run {run.RunId} has neither IME result nor executor end block — dropped ({why})");
|
||||
}
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// IME's own verdict from a run's end block: Success exactly when the error file was empty
|
||||
/// (<c>IsNullOrEmpty(errorFile.Trim())</c> in IME), whatever the exit code — PowerShell returns 0 after a
|
||||
/// non-terminating error and any code a script exits with. Only when the stderr line was never read does the
|
||||
/// exit code stand in.
|
||||
/// </summary>
|
||||
internal static string ImeResultFromEndBlock(ScriptExecutionState script)
|
||||
{
|
||||
if (script.Stderr != null) return string.IsNullOrWhiteSpace(script.Stderr) ? "Success" : "Failed";
|
||||
return script.ExitCode == 0 ? "Success" : "Failed";
|
||||
}
|
||||
|
||||
private void EmitFromExecutorEndBlock(ScriptExecutionState script, string why)
|
||||
{
|
||||
script.Result = ImeResultFromEndBlock(script);
|
||||
script.ResultSource = "agentexecutor_fallback";
|
||||
var basis = script.Stderr == null ? $"exit code {script.ExitCode}, stderr not read" : script.Stderr.Trim().Length == 0 ? "empty stderr" : "stderr present";
|
||||
EmitPlatformScriptResult(script, $"{why} — {script.Result} from the executor end block ({basis}) (fallback)");
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Grace period after a platform script's AgentExecutor exit code is observed before
|
||||
/// <see cref="FlushPendingPlatformScriptResults"/> emits a completion from that exit code.
|
||||
@@ -679,29 +855,48 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// out once its executor end block has been read — the block arrives within one pass, so
|
||||
/// stdout/stderr are in by now — or after <see cref="PlatformScriptEndBlockGrace"/>.</item>
|
||||
/// <item>A script with a known AgentExecutor exit code but no IME <c>PS-SCRIPT-RESULT</c>
|
||||
/// line within <see cref="PlatformScriptResultGrace"/> is emitted from the exit code
|
||||
/// (0 → Success, else Failed), tagged <c>resultSource=agentexecutor_fallback</c> so the
|
||||
/// data is honest about its provenance. Without this, scripts that completed shortly
|
||||
/// before the agent terminates were silently dropped.</item>
|
||||
/// line within <see cref="PlatformScriptResultGrace"/> is emitted from its end block by
|
||||
/// IME's own rule (<see cref="ImeResultFromEndBlock"/>), tagged
|
||||
/// <c>resultSource=agentexecutor_fallback</c> so the data is honest about its provenance.
|
||||
/// Without this, scripts that completed shortly before the agent terminates were silently
|
||||
/// dropped.</item>
|
||||
/// </list>
|
||||
/// Parked runs go first, oldest first, so a policy's runs leave in run order; one without
|
||||
/// result and end block keeps waiting — across a shutdown too — until a later run's result
|
||||
/// resolves it or it is a day old.
|
||||
/// Runs on the single-threaded polling loop (no locking needed, same as the handlers).
|
||||
/// <paramref name="force"/> bypasses both grace checks for the final pass on shutdown.
|
||||
/// </summary>
|
||||
internal void FlushPendingPlatformScriptResults(DateTime nowUtc, bool force = false)
|
||||
{
|
||||
if (_pendingPlatformScripts.Count == 0) return;
|
||||
if (_pendingPlatformScripts.Count == 0 && _parkedPlatformScripts.Count == 0) return;
|
||||
|
||||
var fallbackEmitted = false;
|
||||
foreach (var script in _pendingPlatformScripts.Values.ToList())
|
||||
foreach (var run in _parkedPlatformScripts.OrderBy(StartOf).ToList())
|
||||
{
|
||||
// Already emitted by the authoritative path — just drop the stale buffer entry.
|
||||
if (_platformScriptResultEmitted.ContainsKey(script.PolicyId))
|
||||
if (run.Result != null)
|
||||
{
|
||||
_pendingPlatformScripts.Remove(script.PolicyId);
|
||||
_stateDirty = true;
|
||||
if (run.ExitCode.HasValue || force || !run.ResultHeldSinceUtc.HasValue || nowUtc - run.ResultHeldSinceUtc.Value >= PlatformScriptEndBlockGrace)
|
||||
EmitPlatformScriptResult(run, run.ExitCode.HasValue ? "executor end block read after the result" : $"no executor end block within {PlatformScriptEndBlockGrace.TotalSeconds:F0}s");
|
||||
continue;
|
||||
}
|
||||
if (run.ExitCode.HasValue)
|
||||
{
|
||||
if (!force && (!run.ExitObservedAtUtc.HasValue || nowUtc - run.ExitObservedAtUtc.Value < PlatformScriptResultGrace)) continue;
|
||||
EmitFromExecutorEndBlock(run, $"result not seen within {PlatformScriptResultGrace.TotalSeconds:F0}s{(force ? ", shutdown flush" : string.Empty)}");
|
||||
fallbackEmitted = true;
|
||||
continue;
|
||||
}
|
||||
if (run.StartedAtUtc.HasValue && nowUtc - run.StartedAtUtc.Value > StalePlatformRunAge)
|
||||
{
|
||||
_parkedPlatformScripts.Remove(run);
|
||||
_stateDirty = true;
|
||||
_logger.Debug($"ImeLogTracker: platform script {run.PolicyId} run {run.RunId} waited a day without IME result or executor end block — dropped");
|
||||
}
|
||||
}
|
||||
|
||||
foreach (var script in _pendingPlatformScripts.Values.ToList())
|
||||
{
|
||||
if (script.Result != null)
|
||||
{
|
||||
if (script.ExitCode.HasValue)
|
||||
@@ -720,10 +915,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
if (nowUtc - script.ExitObservedAtUtc.Value < PlatformScriptResultGrace) continue;
|
||||
}
|
||||
|
||||
script.Result = script.ExitCode.Value == 0 ? "Success" : "Failed";
|
||||
script.ResultSource = "agentexecutor_fallback";
|
||||
EmitPlatformScriptResult(script,
|
||||
$"result not seen within {PlatformScriptResultGrace.TotalSeconds:F0}s{(force ? ", shutdown flush" : string.Empty)} — emitted from AgentExecutor exit code {script.ExitCode.Value} (fallback)");
|
||||
EmitFromExecutorEndBlock(script, $"result not seen within {PlatformScriptResultGrace.TotalSeconds:F0}s{(force ? ", shutdown flush" : string.Empty)}");
|
||||
fallbackEmitted = true;
|
||||
}
|
||||
|
||||
@@ -735,7 +927,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// <summary>Test seam: inject a pending platform script as if AgentExecutor.log had reported
|
||||
/// its start + exit code, without an IME PS-SCRIPT-RESULT line.</summary>
|
||||
internal void SeedPendingPlatformScriptForTesting(string policyId, int? exitCode, DateTime? exitObservedAtUtc, string stdout = null, DateTime? startedAtUtc = null,
|
||||
CmTraceLineProvenance startedAtProvenance = null, CmTraceLineProvenance exitProvenance = null)
|
||||
CmTraceLineProvenance startedAtProvenance = null, CmTraceLineProvenance exitProvenance = null, string stderr = null)
|
||||
{
|
||||
// Mirrors a fresh execution start: a new run clears any prior emitted-marker (see
|
||||
// HandleScriptStarted) so re-runs of the same policy within one lifetime emit again.
|
||||
@@ -744,9 +936,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
{
|
||||
PolicyId = policyId,
|
||||
ScriptType = "platform",
|
||||
RunId = NewRunId(),
|
||||
ExitCode = exitCode,
|
||||
ExitObservedAtUtc = exitObservedAtUtc,
|
||||
Stdout = stdout,
|
||||
Stderr = stderr,
|
||||
StartedAtUtc = startedAtUtc,
|
||||
StartedAtProvenance = startedAtProvenance,
|
||||
ExitProvenance = exitProvenance,
|
||||
@@ -755,6 +949,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
RecordInvocationMarker(policyId, isClose: false);
|
||||
}
|
||||
|
||||
/// <summary>Test seam: earlier runs waiting for their IME result while a later run of their policy is in flight.</summary>
|
||||
internal int ParkedPlatformScriptCountForTest => _parkedPlatformScripts.Count;
|
||||
|
||||
/// <summary>Test seam: the authoritative IME PS-SCRIPT-RESULT path for a platform script
|
||||
/// (the production handler minus regex extraction). <paramref name="resultLineTimestampUtc"/>
|
||||
/// stands in for the CMTrace timestamp of the PS-SCRIPT-RESULT line (production reads it
|
||||
|
||||
+11
-6
@@ -123,8 +123,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
var ledger = IsMultiWriterLogFile(fileName) ? GetLedger(fileName) : null;
|
||||
|
||||
// Set by a rewind: entries that still read unchanged behind the old bookmark are
|
||||
// skipped (their invocation markers are given back), and nothing below the old
|
||||
// bookmark is fresh or a calibration anchor.
|
||||
// skipped (their invocation markers are given back); a changed entry below the old
|
||||
// bookmark is fresh only when the bytes it replaced were seen moments ago, and never a
|
||||
// calibration anchor.
|
||||
RewindState rewind = null;
|
||||
long oldBookmark = -1;
|
||||
var writerBoundaries = IsWriterBoundaryLogFile(fileName);
|
||||
@@ -171,12 +172,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
var guardFrom = ledger.OffsetOfEntryAtOrAfter(startPos - OverwriteGuardBytes);
|
||||
if (guardFrom >= 0 && guardFrom < startPos)
|
||||
{
|
||||
var divergence = await FindOverwriteAsync(stream, ledger, guardFrom, startPos, token);
|
||||
var divergence = await FindOverwriteAsync(stream, ledger, guardFrom, startPos, passNowUtc, token);
|
||||
_verifiedBytes += startPos - guardFrom;
|
||||
if (divergence >= 0)
|
||||
{
|
||||
// The guard saw only the tail of the ledger; the block may start earlier.
|
||||
var earliest = await FindOverwriteAsync(stream, ledger, ledger.StartOffset, startPos, token);
|
||||
var earliest = await FindOverwriteAsync(stream, ledger, ledger.StartOffset, startPos, passNowUtc, token);
|
||||
_verifiedBytes += startPos - ledger.StartOffset;
|
||||
_verifyPasses++;
|
||||
if (earliest >= 0) divergence = earliest;
|
||||
@@ -271,7 +272,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
}
|
||||
}
|
||||
|
||||
if (ledger != null) ledger.Record(entry);
|
||||
if (ledger != null) ledger.Record(entry, passNowUtc);
|
||||
var unchanged = rewind != null && rewind.IsUnchanged(entry);
|
||||
if (unchanged) RestoreInvocationMarkers(rewind, entry.Offset);
|
||||
|
||||
@@ -287,7 +288,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
|
||||
if (unchanged) continue; // unchanged behind the old bookmark — processed before the rewind
|
||||
|
||||
_currentPassLinesAreFresh = passLinesAreFresh && entry.Offset >= oldBookmark;
|
||||
// Behind the old bookmark only changed entries arrive here: fresh when the bytes they replaced
|
||||
// were seen moments ago (RewoundEntryIsFresh), never a calibration anchor.
|
||||
_currentPassLinesAreFresh = entry.Offset >= oldBookmark
|
||||
? passLinesAreFresh
|
||||
: rewind != null && RewoundEntryIsFresh(passLinesAreFresh, rewind.EntryAtRewindPointReplaced, rewind.OldestReplacedSeenUtc, passNowUtc);
|
||||
|
||||
// --- Normal processing (single-line or completed multiline) ---
|
||||
CmTraceLogEntry parsed;
|
||||
|
||||
+55
-11
@@ -62,6 +62,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
public long Offset;
|
||||
public int Length;
|
||||
public ulong Hash;
|
||||
/// <summary>
|
||||
/// When these bytes were last seen as they are (read, seeded or verified unchanged): bytes found different
|
||||
/// at this position later were written after it. Not part of the identity.
|
||||
/// </summary>
|
||||
public DateTime SeenUtc;
|
||||
|
||||
// Identity is (offset, hash): a tail first read unterminated and later completed
|
||||
// keeps its hash but grows by its terminator.
|
||||
@@ -100,11 +105,11 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
return idx < Entries.Count ? Entries[idx].Offset : -1;
|
||||
}
|
||||
|
||||
public void Record(AssembledEntry e)
|
||||
public void Record(AssembledEntry e, DateTime seenUtc)
|
||||
{
|
||||
var idx = IndexOfFirstAtOrAfter(e.Offset);
|
||||
if (idx < Entries.Count) Entries.RemoveRange(idx, Entries.Count - idx);
|
||||
Entries.Add(new LedgerEntry { Offset = e.Offset, Length = e.Length, Hash = e.Hash });
|
||||
Entries.Add(new LedgerEntry { Offset = e.Offset, Length = e.Length, Hash = e.Hash, SeenUtc = seenUtc });
|
||||
}
|
||||
|
||||
/// <summary>Drops everything from <paramref name="offset"/> on and returns it — the entries a rewind must not process twice.</summary>
|
||||
@@ -197,7 +202,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
{
|
||||
ledger.Seeded = true;
|
||||
if (ledger.Entries.Count == 0 && bookmark > 0)
|
||||
await SeedLedgerAsync(filePath, ledger, bookmark, token).ConfigureAwait(false);
|
||||
await SeedLedgerAsync(filePath, ledger, bookmark, nowUtc, token).ConfigureAwait(false);
|
||||
}
|
||||
|
||||
var due = ledger.VerifyRequested || (InContention(nowUtc) && nowUtc >= ledger.NextCadenceVerifyUtc);
|
||||
@@ -213,7 +218,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
|
||||
long divergence;
|
||||
using (var stream = new FileStream(filePath, FileMode.Open, FileAccess.Read, FileShare.ReadWrite))
|
||||
divergence = await FindOverwriteAsync(stream, ledger, from, bookmark, token).ConfigureAwait(false);
|
||||
divergence = await FindOverwriteAsync(stream, ledger, from, bookmark, nowUtc, token).ConfigureAwait(false);
|
||||
if (token.IsCancellationRequested) return -1;
|
||||
|
||||
ledger.VerifyRequested = false;
|
||||
@@ -228,8 +233,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// Re-reads [fromOffset, toOffset) with the read loop's assembler and compares against the
|
||||
/// ledger. Returns the offset of the first entry whose bytes changed — a ledger entry not
|
||||
/// reproduced, or an entry the ledger never had — or -1 when everything reads as before.
|
||||
/// Every entry found unchanged is stamped seen at <paramref name="nowUtc"/>.
|
||||
/// </summary>
|
||||
private async Task<long> FindOverwriteAsync(FileStream stream, FileLedger ledger, long fromOffset, long toOffset, CancellationToken token)
|
||||
private async Task<long> FindOverwriteAsync(FileStream stream, FileLedger ledger, long fromOffset, long toOffset, DateTime nowUtc, CancellationToken token)
|
||||
{
|
||||
var entries = ledger.Entries;
|
||||
var idx = ledger.IndexOfFirstAtOrAfter(fromOffset);
|
||||
@@ -255,6 +261,9 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
if (idx < entries.Count && entries[idx].Offset == e.Offset)
|
||||
{
|
||||
if (entries[idx].Hash != e.Hash) return e.Offset;
|
||||
var confirmed = entries[idx];
|
||||
confirmed.SeenUtc = nowUtc;
|
||||
entries[idx] = confirmed;
|
||||
idx++;
|
||||
continue;
|
||||
}
|
||||
@@ -265,7 +274,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
}
|
||||
|
||||
/// <summary>After a restart the ledger is empty: fingerprint the entries just behind the persisted bookmark without processing them.</summary>
|
||||
private async Task SeedLedgerAsync(string filePath, FileLedger ledger, long bookmark, CancellationToken token)
|
||||
private async Task SeedLedgerAsync(string filePath, FileLedger ledger, long bookmark, DateTime nowUtc, CancellationToken token)
|
||||
{
|
||||
var from = Math.Max(0, bookmark - RestartSeedBytes);
|
||||
using (var stream = new FileStream(filePath, FileMode.Open, FileAccess.Read, FileShare.ReadWrite))
|
||||
@@ -288,7 +297,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
if (assembler.Feed(line, reader.LastLineStart, reader.Position, reader.LastLineTruncated, reader.LastLineHash, out e)
|
||||
== CmTraceEntryAssembler.Outcome.Entry
|
||||
&& e.Offset + e.Length <= bookmark)
|
||||
ledger.Record(e);
|
||||
ledger.Record(e, nowUtc);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -300,12 +309,17 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// </summary>
|
||||
private sealed class RewindState
|
||||
{
|
||||
public RewindState(string fileName, HashSet<LedgerEntry> unchanged, List<InvocationMarker> markers, InvocationMarker lastPlatformMarker)
|
||||
public RewindState(string fileName, HashSet<LedgerEntry> unchanged, List<InvocationMarker> markers, InvocationMarker lastPlatformMarker, long rewindTo)
|
||||
{
|
||||
FileName = fileName;
|
||||
Unchanged = unchanged;
|
||||
Markers = markers;
|
||||
LastPlatformMarker = lastPlatformMarker;
|
||||
foreach (var replaced in unchanged)
|
||||
{
|
||||
if (replaced.Offset == rewindTo) EntryAtRewindPointReplaced = true;
|
||||
if (!OldestReplacedSeenUtc.HasValue || replaced.SeenUtc < OldestReplacedSeenUtc.Value) OldestReplacedSeenUtc = replaced.SeenUtc;
|
||||
}
|
||||
}
|
||||
|
||||
public string FileName { get; }
|
||||
@@ -315,10 +329,29 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// <summary>The tracker-wide last platform marker, when it was among the set-aside ones.</summary>
|
||||
public InvocationMarker LastPlatformMarker { get; }
|
||||
public int Next { get; set; }
|
||||
/// <summary>The ledger held the entry at the rewind point: the changed bytes lie where something was seen before.</summary>
|
||||
public bool EntryAtRewindPointReplaced { get; }
|
||||
/// <summary>The earliest time any of the replaced bytes was last seen as it was: every changed byte was written after it.</summary>
|
||||
public DateTime? OldestReplacedSeenUtc { get; }
|
||||
|
||||
public bool IsUnchanged(AssembledEntry entry) => Unchanged.Contains(new LedgerEntry { Offset = entry.Offset, Hash = entry.Hash });
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Whether an entry a rewind re-processes (it reads differently than before) is fresh. Its bytes were written after
|
||||
/// the bytes it replaced were last seen as they were — IME stamps a line when it writes it — so when that was at
|
||||
/// most <see cref="FreshLineMaxAge"/> ago the entry's age is bounded like an appended line's and it may anchor
|
||||
/// itself (session d29e27e3: recovered result lines fell back to the agent's stale zone and lost their duration).
|
||||
/// Not fresh when the pass is not (the first pass after a restart never anchors), when the divergence lies in bytes
|
||||
/// the ledger never held (nothing dates them), or when the clock stepped back since.
|
||||
/// </summary>
|
||||
internal static bool RewoundEntryIsFresh(bool passLinesAreFresh, bool entryAtRewindPointReplaced, DateTime? oldestReplacedSeenUtc, DateTime passNowUtc)
|
||||
{
|
||||
if (!passLinesAreFresh || !entryAtRewindPointReplaced || !oldestReplacedSeenUtc.HasValue) return false;
|
||||
var sinceSeen = passNowUtc - oldestReplacedSeenUtc.Value;
|
||||
return sinceSeen >= TimeSpan.Zero && sinceSeen <= FreshLineMaxAge;
|
||||
}
|
||||
|
||||
/// <summary>Moves the bookmark back to the first changed entry; returns what the re-read needs to skip and to give back.</summary>
|
||||
private RewindState ApplyRewind(string filePath, string fileName, FileLedger ledger, long rewindTo, long bookmark)
|
||||
{
|
||||
@@ -341,7 +374,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
{
|
||||
_logger.Debug(message);
|
||||
}
|
||||
return new RewindState(fileName, removed, markers, tookLastPlatformMarker ? lastPlatformMarker : null);
|
||||
return new RewindState(fileName, removed, markers, tookLastPlatformMarker ? lastPlatformMarker : null, rewindTo);
|
||||
}
|
||||
|
||||
private long RetentionFloor(string fileName, long bookmark)
|
||||
@@ -380,11 +413,20 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
{
|
||||
public long Offset;
|
||||
public string PolicyId;
|
||||
/// <summary>
|
||||
/// The platform run the start line belongs to: the lines behind it go to that run while it is pending or
|
||||
/// waits for its IME result, and nowhere once it was emitted — never to a later run of the same policy.
|
||||
/// Null only on seeded test markers, which resolve by policy.
|
||||
/// </summary>
|
||||
public string RunId;
|
||||
public bool IsClose;
|
||||
/// <summary>Set on a writer boundary: the platform invocation it closed.</summary>
|
||||
public string ClosedPolicyId;
|
||||
}
|
||||
|
||||
/// <summary>Run id of a start line whose run was emitted already (a late line): owns its lines, matches no live run.</summary>
|
||||
private const string EmittedRunId = "emitted";
|
||||
|
||||
private const string ExecutorLogFileName = "AgentExecutor.log";
|
||||
|
||||
private readonly Dictionary<string, List<InvocationMarker>> _invocationMarkers =
|
||||
@@ -437,7 +479,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
list.Add(marker);
|
||||
}
|
||||
|
||||
private void RecordInvocationMarker(string policyId, bool isClose)
|
||||
private void RecordInvocationMarker(string policyId, bool isClose, string runId = null)
|
||||
{
|
||||
if (_currentEntryOffset < 0) return;
|
||||
var file = _currentSourceFileName ?? TestSourceFileName;
|
||||
@@ -451,6 +493,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
if (!isClose && policyId != null && last.ClosedPolicyId == null)
|
||||
{
|
||||
last.PolicyId = policyId;
|
||||
last.RunId = runId;
|
||||
last.IsClose = false;
|
||||
_lastPlatformMarker = last;
|
||||
_platformMarkerFiles.Add(file);
|
||||
@@ -458,7 +501,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
return;
|
||||
}
|
||||
|
||||
var marker = new InvocationMarker { Offset = _currentEntryOffset, PolicyId = policyId, IsClose = isClose };
|
||||
var marker = new InvocationMarker { Offset = _currentEntryOffset, PolicyId = policyId, RunId = runId, IsClose = isClose };
|
||||
AppendInvocationMarker(list, marker);
|
||||
if (!isClose && policyId != null)
|
||||
{
|
||||
@@ -526,6 +569,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
if (marker == null || marker.IsClose || marker.PolicyId == null) return null;
|
||||
|
||||
ownerPolicyId = marker.PolicyId;
|
||||
if (marker.RunId != null) return FindLivePlatformRun(marker.PolicyId, marker.RunId);
|
||||
ScriptExecutionState state;
|
||||
return _pendingPlatformScripts.TryGetValue(marker.PolicyId, out state) ? state : null;
|
||||
}
|
||||
|
||||
@@ -182,6 +182,13 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
private readonly Dictionary<string, ScriptExecutionState> _pendingPlatformScripts =
|
||||
new Dictionary<string, ScriptExecutionState>(StringComparer.OrdinalIgnoreCase);
|
||||
|
||||
// Earlier runs of a policy whose next executor run started before IME's result of them was
|
||||
// read (IME re-runs a failed script on later check-ins; the first pass reads all of
|
||||
// AgentExecutor.log before the IME log). Each keeps its own end block until its result
|
||||
// arrives — session eff1b413 lost that and paired the first result with the last run's block.
|
||||
private readonly List<ScriptExecutionState> _parkedPlatformScripts = new List<ScriptExecutionState>();
|
||||
private const int MaxParkedPlatformScripts = 64;
|
||||
|
||||
/// <summary>
|
||||
/// The platform runs this tracker started and emitted, for the registry reconciliation that
|
||||
/// checks them against IME's saved results (D-316). Shared with the registry host's thread.
|
||||
@@ -1191,6 +1198,16 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
}
|
||||
}
|
||||
|
||||
_parkedPlatformScripts.Clear();
|
||||
if (state.ParkedPlatformScripts != null)
|
||||
{
|
||||
foreach (var parked in state.ParkedPlatformScripts)
|
||||
{
|
||||
if (!string.IsNullOrEmpty(parked?.PolicyId) && _parkedPlatformScripts.Count < MaxParkedPlatformScripts)
|
||||
_parkedPlatformScripts.Add(parked);
|
||||
}
|
||||
}
|
||||
|
||||
_scriptTimeoutSuspectedPosted.Clear();
|
||||
if (state.ScriptTimeoutSuspectedPosted != null)
|
||||
{
|
||||
@@ -1276,6 +1293,7 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
// must survive restarts — see the LoadState comment for the duplicate scenario.
|
||||
PlatformScriptResultEmittedAt = new Dictionary<string, DateTime>(_platformScriptResultEmitted, StringComparer.OrdinalIgnoreCase),
|
||||
PendingPlatformScripts = _pendingPlatformScripts.Values.ToList(),
|
||||
ParkedPlatformScripts = _parkedPlatformScripts.ToList(),
|
||||
ScriptTimeoutSuspectedPosted = _scriptTimeoutSuspectedPosted.ToList(),
|
||||
RecurringScripts = _recurringScripts.ToPersisted(),
|
||||
};
|
||||
|
||||
+4
@@ -151,6 +151,10 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
// that run from the start of the next one.
|
||||
public Dictionary<string, DateTime> PlatformScriptResultEmittedAt { get; set; }
|
||||
public List<ScriptExecutionState> PendingPlatformScripts { get; set; }
|
||||
// Earlier runs that wait for their IME result while a later run of the same policy is in
|
||||
// flight; a restart between reading AgentExecutor.log and the IME log must keep them. Null on
|
||||
// older state files → none.
|
||||
public List<ScriptExecutionState> ParkedPlatformScripts { get; set; }
|
||||
public List<string> ScriptTimeoutSuspectedPosted { get; set; }
|
||||
|
||||
// Recurring health-script runs collapsed per policy. Saved in the same write as
|
||||
|
||||
+6
-3
@@ -41,9 +41,12 @@ namespace AutopilotMonitor.Agent.V2.Core.Monitoring.Enrollment.Ime
|
||||
/// Written on the tracker's poll thread, claimed on the registry host's timer thread: every member
|
||||
/// locks, and nothing calls out while holding the lock.
|
||||
/// <para>
|
||||
/// A save belongs to the latest run of its key: IME runs a policy once per batch and saves the
|
||||
/// batch after its last script, so the next run of the same key starts minutes later at the
|
||||
/// earliest. A save is paired only after the tracker completed a pass that started
|
||||
/// A save belongs to the latest run of its key: IME runs a policy at most once per check-in and
|
||||
/// saves after reporting that check-in's batch, so the next run of the key comes with a later
|
||||
/// check-in. (Overlapping check-ins can run a key twice at once; then IME itself shares the key's
|
||||
/// result files and the judge's foreign verdict is the safety net.) A run that waited for its IME
|
||||
/// result while a later run of its key already started is emitted without touching that later
|
||||
/// run's entry. A save is paired only after the tracker completed a pass that started
|
||||
/// <see cref="TrackerCatchUp"/> after the save was seen — every line written before the save, the
|
||||
/// run's start among them, has been read by then, a stalled poll loop included.
|
||||
/// </para>
|
||||
|
||||
+11
-2
@@ -23,11 +23,20 @@ public class ScriptExecutionState
|
||||
public string UserId { get; set; }
|
||||
|
||||
/// <summary>
|
||||
/// Identity of one emitted platform-script run, surfaced as <c>runId</c> on the event: what a
|
||||
/// later <c>script_output_reconciliation</c> names when it corrects that run.
|
||||
/// Identity of one platform-script run, given when its slot opens and surfaced as <c>runId</c> on
|
||||
/// the event: the invocation markers bind the run's executor lines to it, and a later
|
||||
/// <c>script_output_reconciliation</c> names it when it corrects that run.
|
||||
/// </summary>
|
||||
public string RunId { get; set; }
|
||||
|
||||
/// <summary>
|
||||
/// Source timestamp of the run's AgentExecutor start line (<c>Adding argument powershell with
|
||||
/// value …</c>). Every executor process writes that line once, so a later one for the same
|
||||
/// policy starts the next run, never this one. Null until the line is read, and on state files
|
||||
/// from before the field.
|
||||
/// </summary>
|
||||
public System.DateTime? ExecutorStartedAtUtc { get; set; }
|
||||
|
||||
/// <summary>
|
||||
/// UTC timestamp of the script's first observed start line, taken from the source CMTrace log
|
||||
/// timestamp (so it dates correctly even when the agent replays historic IME log content that
|
||||
|
||||
Reference in New Issue
Block a user