You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@ignite.apache.org by rp...@apache.org on 2025/04/29 15:25:35 UTC
(ignite-3) branch main updated: IGNITE-24876 Fix message reordering in network (#5723)
This is an automated email from the ASF dual-hosted git repository.
rpuch pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/ignite-3.git
The following commit(s) were added to refs/heads/main by this push:
new a7a4b679268 IGNITE-24876 Fix message reordering in network (#5723)
a7a4b679268 is described below
commit a7a4b67926842ca66215529a12fb72bc827fe1d2
Author: Roman Puchkovskiy <ro...@gmail.com>
AuthorDate: Tue Apr 29 19:25:30 2025 +0400
IGNITE-24876 Fix message reordering in network (#5723)
The reordering was happening because messages waiting for a sender to be opened (in the sender awaiting queue) would be written directly to the channel (when the sender was finally opened), bypassing the event loop queue.
---
.../org/apache/ignite/internal/network/netty/NettySender.java | 8 +++-----
1 file changed, 3 insertions(+), 5 deletions(-)
diff --git a/modules/network/src/main/java/org/apache/ignite/internal/network/netty/NettySender.java b/modules/network/src/main/java/org/apache/ignite/internal/network/netty/NettySender.java
index 9961ec4833c..6c11772a7d6 100644
--- a/modules/network/src/main/java/org/apache/ignite/internal/network/netty/NettySender.java
+++ b/modules/network/src/main/java/org/apache/ignite/internal/network/netty/NettySender.java
@@ -108,11 +108,9 @@ public class NettySender {
// Write in event loop to make sure that, if a ClosedSocketException happens, we recover from it without exiting the event loop.
// We need this to avoid message reordering due to switching from old channel to a new one.
- if (channel.eventLoop().inEventLoop()) {
- writeWithRecovery(obj, channel, triggerChannelRecreation);
- } else {
- channel.eventLoop().execute(() -> writeWithRecovery(obj, channel, triggerChannelRecreation));
- }
+ // Also, we ALWAYS execute the writes on the event loop (by adding them to the event loop queue) even if we are on
+ // the event loop thread, to avoid another reordering.
+ channel.eventLoop().execute(() -> writeWithRecovery(obj, channel, triggerChannelRecreation));
return obj.acknowledgedFuture();
}