RPC performance changes

- configurable ping timeout
- single, much higher, configurable RPC timeout
- no more concurrency semaphore
This commit is contained in:
Alex Auvolat
2022-09-19 20:12:19 +02:00
parent fbd32933ea
commit 56592e1853
13 changed files with 99 additions and 138 deletions
+2 -8
View File
@@ -25,8 +25,6 @@ use crate::replication::*;
use crate::schema::*;
const TABLE_GC_BATCH_SIZE: usize = 1024;
// Same timeout as NEED_BLOCK_QUERY_TIMEOUT in block manager
const TABLE_GC_RPC_TIMEOUT: Duration = Duration::from_secs(15);
// GC delay for table entries: 1 day (24 hours)
// (the delay before the entry is added in the GC todo list
@@ -237,9 +235,7 @@ where
&self.endpoint,
&nodes[..],
GcRpc::Update(updates),
RequestStrategy::with_priority(PRIO_BACKGROUND)
.with_quorum(nodes.len())
.with_timeout(TABLE_GC_RPC_TIMEOUT),
RequestStrategy::with_priority(PRIO_BACKGROUND).with_quorum(nodes.len()),
)
.await
.err_context("GC: send tombstones")?;
@@ -260,9 +256,7 @@ where
&self.endpoint,
&nodes[..],
GcRpc::DeleteIfEqualHash(deletes),
RequestStrategy::with_priority(PRIO_BACKGROUND)
.with_quorum(nodes.len())
.with_timeout(TABLE_GC_RPC_TIMEOUT),
RequestStrategy::with_priority(PRIO_BACKGROUND).with_quorum(nodes.len()),
)
.await
.err_context("GC: remote delete tombstones")?;