diff options
| author | tomyouyou <yxszyn@163.com> | 2019-09-04 11:51:08 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2019-09-04 11:51:08 +0800 |
| commit | dc2552b682633660b75e8b945398f4d6f29ccc32 (patch) | |
| tree | 924fd9e63c79c4819cc3e4a1b9bd1050b861b806 | |
| parent | 133d261ad615d377060c2b6ca0deaa262f109ed9 (diff) | |
| download | rabbitmq-server-git-dc2552b682633660b75e8b945398f4d6f29ccc32.tar.gz | |
call "add_to_journal" before "maybe_flush_pre_publish_cache"
To reproduce the issue:
1. pre_publish a delivered message.
2. Put its SeqId into delivered_cache, and might call "flush_pre_publish_cache" and "flush_delivered_cache",
then put its "del" data into the journal_entries of the corresponding segment.
3. Call add_to_journal and action_to_entry with a "?PUB" Action.
4. in action_to_entry, "array:get(RelSeq, JEntries)" will return {no_pub, del, no_ack}, but has none case for a "?PUB" Action, so an error will occur like "{{case_clause,{no_pub,del,no_ack}},[{rabbit_queue_index,action_to_entry,3..."
| -rw-r--r-- | src/rabbit_queue_index.erl | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/rabbit_queue_index.erl b/src/rabbit_queue_index.erl index 2805584124..e4f31ab36f 100644 --- a/src/rabbit_queue_index.erl +++ b/src/rabbit_queue_index.erl @@ -349,11 +349,11 @@ pre_publish(MsgOrId, SeqId, MsgProps, IsPersistent, IsDelivered, JournalSizeHint DC end, - add_to_journal(SeqId, {IsPersistent, Bin, MsgBin}, - maybe_flush_pre_publish_cache( + State2 = add_to_journal(SeqId, {IsPersistent, Bin, MsgBin}, State1), + maybe_flush_pre_publish_cache( JournalSizeHint, - State1#qistate{pre_publish_cache = PPC1, - delivered_cache = DC1})). + State2#qistate{pre_publish_cache = PPC1, + delivered_cache = DC1}). %% pre_publish_cache is the entry with most elements when compared to %% delivered_cache so we only check the former in the guard. |