(kudu) branch master updated: [consensus] Fix heartbeat batching upper limit.

[alexey]

This is an automated email from the ASF dual-hosted git repository.

alexey pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/kudu.git


The following commit(s) were added to refs/heads/master by this push:
     new 5c07add3c [consensus] Fix heartbeat batching upper limit.
5c07add3c is described below

commit 5c07add3c06a23a7333ff87daccd1ac79ae7d855
Author: Zoltan Martonka <zmarto...@gmail.com>
AuthorDate: Mon Feb 9 09:01:54 2026 +0000

    [consensus] Fix heartbeat batching upper limit.
    
    multi_raft_heartbeat_window_ms should be no more than
    raft_heartbeat_interval_ms / 2.
    Currently it is capped at FLAGS_consensus_rpc_timeout_ms / 2,
    which is simply wrong.
    
    Change-Id: Ib52f21238c7c9e76828edf5582f12b063234f56b
    Reviewed-on: http://gerrit.cloudera.org:8080/23952
    Tested-by: Kudu Jenkins
    Reviewed-by: Zoltan Chovan <zcho...@cloudera.com>
    Reviewed-by: Gabriella Lotz <lotzgabrie...@gmail.com>
    Reviewed-by: Alexey Serbin <ale...@apache.org>
---
 src/kudu/consensus/multi_raft_batcher.cc | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/src/kudu/consensus/multi_raft_batcher.cc 
b/src/kudu/consensus/multi_raft_batcher.cc
index 87c23023a..933451a13 100644
--- a/src/kudu/consensus/multi_raft_batcher.cc
+++ b/src/kudu/consensus/multi_raft_batcher.cc
@@ -58,12 +58,14 @@ DECLARE_int32(raft_heartbeat_interval_ms);
 DEFINE_int32(multi_raft_heartbeat_window_ms, 100,
   "The batch time window for heartbeat batching. "
   "For minimal delay and still maximum effectiveness set "
-  "multi_raft_heartbeat_window_ms = heartbeat_interval_ms * ("
-  "batch_size / {estimated follower peers from the same host}) + {a little 
tolerance}; "
-  "however, a 0.1 * heartbeat_interval_ms is good, and there is no reason to "
+  "multi_raft_heartbeat_window_ms = raft_heartbeat_interval_ms * ("
+  "multi_raft_batch_size / {estimated follower peers from the same host})"
+  " + {a little tolerance}; "
+  "however, a 0.1 * raft_heartbeat_interval_ms is good, and there is no reason 
to "
   "set it any lower. "
-  "This value is also forced to be less than or equal to half of the heartbeat 
interval, "
-  "because it makes no sense to introduce a possible delay comparable to the 
heartbeat interval."
+  "This value is also forced to be less than or equal to half of "
+  "raft_heartbeat_interval_ms, because it makes no sense to introduce a 
possible "
+  "delay comparable to the heartbeat interval."
 );
 TAG_FLAG(multi_raft_heartbeat_window_ms, experimental);
 
@@ -224,14 +226,14 @@ void MultiRaftHeartbeatBatcher::SendOutScheduled(
 
 namespace {
 MonoDelta GetTimeWindow() {
-  // We don't want to delay more than half of the heartbeat interval.
+  // We don't want to delay more than half of the raft heartbeat interval.
   // Even a quarter of the interval starts to be questionable delay,
   // but half is for sure too much.
   auto flush_interval =
-      std::min(FLAGS_multi_raft_heartbeat_window_ms, 
FLAGS_consensus_rpc_timeout_ms / 2);
+      std::min(FLAGS_multi_raft_heartbeat_window_ms, 
FLAGS_raft_heartbeat_interval_ms / 2);
   if (flush_interval != FLAGS_multi_raft_heartbeat_window_ms) {
     LOG(ERROR) << "multi_raft_heartbeat_window_ms should not be more than "
-               << " consensus_rpc_timeout_ms / 2. , forcing 
multi_raft_heartbeat_window_ms = "
+               << " raft_heartbeat_interval_ms / 2. , forcing 
multi_raft_heartbeat_window_ms = "
                << flush_interval;
   }
   return MonoDelta::FromMilliseconds(flush_interval);