This is an automated email from the ASF dual-hosted git repository.
bobhan1 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/doris.git
The following commit(s) were added to refs/heads/master by this push:
new 2763903be96 [fix](be) Fix WAL replay retry backoff overflow (#68834)
2763903be96 is described below
commit 2763903be964e3a814278001d0425c5a5f30d24d
Author: bobhan1 <[email protected]>
AuthorDate: Sat Oct 10 10:06:17 2026 +0800
[fix](be) Fix WAL replay retry backoff overflow (#68834)
### What problem does this PR solve?
Issue Number: N/A
Related PR: N/A
Problem Summary: Long-running WAL replay retries can lose their backoff
because the linear delay is multiplied using 32-bit integers before the
result is converted to `int64_t`. With the default configuration, retry
count 1204 makes `(1204 - 10) * 1800 * 1000` exceed `INT32_MAX`,
potentially causing replay on every background scan instead of every 30
minutes.
Promote the retry count to `int64_t` before subtraction and
multiplication. This preserves the existing backoff formula while
keeping the linear delay in 64-bit arithmetic.
Move the `BE_TEST` bypass to the start of `WalTable::_need_replay()` so
test builds return immediately before checking replay configuration,
preserving immediate replay in test builds.
### Release note
Fix WAL replay backoff overflow after prolonged retries.
### Check List (For Author)
- Test:
- [x] No need to test or manual test. Explain why:
- [x] Other reason: Small arithmetic type promotion and a
behavior-preserving cleanup of the test guard. No tests added at the
user's request. No automated tests or full build run. No further
clang-tidy runs at the user's request; the initial attempt could not
analyze the file because the local toolchain could not find
`emmintrin.h`.
- Validation: `build-support/clang-format.sh` and
`build-support/check-format.sh` passed with clang-format 16.0.6; `git
diff --check` passed. Manually checked the default delay at retry counts
1203, 1204, and 1205: 2,152,520,000, 2,154,320,000, and 2,156,120,000
ms, respectively, retaining a 1,800,000 ms increment.
- Behavior changed:
- [x] Yes. Prolonged failed WAL replays retain the configured backoff
after the previous 32-bit overflow boundary.
- Does this need documentation?
- [x] No.
### Check List (For Reviewer who merge this PR)
- [ ] Confirm the release note
- [ ] Confirm test cases
- [ ] Confirm document
- [ ] Add branch pick label
---
be/src/load/group_commit/wal/wal_table.cpp | 17 ++++++++---------
1 file changed, 8 insertions(+), 9 deletions(-)
diff --git a/be/src/load/group_commit/wal/wal_table.cpp
b/be/src/load/group_commit/wal/wal_table.cpp
index 69e5d7cf2b7..2b7db81b664 100644
--- a/be/src/load/group_commit/wal/wal_table.cpp
+++ b/be/src/load/group_commit/wal/wal_table.cpp
@@ -157,25 +157,24 @@ Status WalTable::replay_wals() {
}
bool WalTable::_need_replay(std::shared_ptr<WalInfo> wal_info) {
+#ifdef BE_TEST
+ return true;
+#endif
if (config::group_commit_wait_replay_wal_finish) {
return true;
}
-#ifndef BE_TEST
int64_t replay_interval = 0;
if (wal_info->get_retry_num() >=
config::group_commit_replay_wal_retry_num) {
- replay_interval =
- int64_t(pow(2, config::group_commit_replay_wal_retry_num) *
-
config::group_commit_replay_wal_retry_interval_seconds * 1000 +
- (wal_info->get_retry_num() -
config::group_commit_replay_wal_retry_num) *
-
config::group_commit_replay_wal_retry_interval_max_seconds * 1000);
+ replay_interval = int64_t(
+ pow(2, config::group_commit_replay_wal_retry_num) *
+ config::group_commit_replay_wal_retry_interval_seconds
* 1000 +
+ (int64_t(wal_info->get_retry_num()) -
config::group_commit_replay_wal_retry_num) *
+
config::group_commit_replay_wal_retry_interval_max_seconds * 1000);
} else {
replay_interval = int64_t(pow(2, wal_info->get_retry_num()) *
config::group_commit_replay_wal_retry_interval_seconds * 1000);
}
return UnixMillis() - wal_info->get_start_time_ms() >= replay_interval;
-#else
- return true;
-#endif
}
Status WalTable::_try_abort_txn(int64_t db_id, std::string& label) {
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
