gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-30 16:59:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf(ecstore): improve erasure write diagnostics and single-block performance (#3280)

Browse Source 
* docs(object-capacity): add localized crate docs

* fix(ecstore): improve quorum and transport diagnostics

* perf(ecstore): add safe single-block write fast path

* refactor(ecstore): collapse layered small write paths

* chore(docs): keep issue 662 design note tracked

* fix(docs): restore issue 662 design note

* chore(docs): keep issue 662 design local only

* feat(obs): add internode reliability metrics and dashboard

* feat(obs): extend internode diagnostics and service logging

* fix(docs): use AGENTS guide filename

* perf(ecstore): reuse owned buffer in small encode

* fix(ecstore): tighten small write diagnostics

---------

Co-authored-by: cxymds <Cxymds@qq.com>
This commit is contained in:
houseme
2026-06-08 17:45:56 +08:00
committed by GitHub
parent ddd35badad
commit ed73952cb6
16 changed files with 2455 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/io-metrics/README.md
+19
View File
@@ -167,6 +167,10 @@ Operation-level metrics use the same low-cardinality label set:
| `rustfs_system_network_internode_operation_requests_outgoing_total` | `operation`, `backend` | Outgoing request attempts for an internode operation. |
| `rustfs_system_network_internode_operation_requests_incoming_total` | `operation`, `backend` | Incoming request attempts for an internode operation. |
| `rustfs_system_network_internode_operation_errors_total` | `operation`, `backend` | Failed internode operation attempts. |
| `rustfs_system_network_internode_operation_classified_errors_total` | `operation`, `backend`, `classification` | Classified internode transport failures. |
| `rustfs_system_network_internode_operation_retries_total` | `operation`, `backend`, `classification` | Retry attempts for retryable internode transport failures. |
| `rustfs_system_network_internode_operation_retry_successes_total` | `operation`, `backend`, `classification` | Successful recoveries after retryable internode transport failures. |
| `rustfs_system_storage_erasure_write_quorum_failures_total` | `stage`, `dominant_error` | Erasure write quorum failures grouped by failure stage and dominant error class. |
Current `operation` values are `read_file_stream`, `put_file_stream`,
`walk_dir`, `grpc_read_all`, and `grpc_write_all`. Current `backend` values are
@@ -180,6 +184,21 @@ successful completions are not emitted as a dedicated result-labeled metric.
Adding completion/result labels is a follow-up once stream completion semantics
are defined consistently for request setup, body transfer, and shutdown.
Current low-cardinality `classification` values come from the TCP/HTTP internode
path and include:
- `connect_timeout`
- `connection_refused`
- `dns_resolution_failed`
- `connection_reset`
- `body_stream_aborted`
- `http_429`
- `http_502`
- `http_503`
- `http_504`
- `http_status_other`
- `unknown`
`scripts/run_internode_transport_baseline.sh --metrics-url ...` records metric
deltas with `operation` and `backend` columns, so the TCP baseline can attribute
bytes and request/error counts to `tcp-http` transport operations.