bobhan1 opened a new pull request, #70653: URL: https://github.com/apache/doris/pull/70653
### What problem does this PR solve? Related PR: #70478 Backport commit `5e85976d4c58d5528cb49d3ef677d5e695c2a36d` to `branch-4.1`. A single locked delete bitmap can exceed the FoundationDB transaction limit even though its value is split into 90 KB KVs. Split oversized v1/v2 values across metadata transactions using the existing transaction budget, and verify the committed prefix in the last data transaction with non-snapshot reads. Missing or replaced prefix fragments fail with `KV_TXN_CONFLICT`; unlocked aggregation retains atomic overwrites. The conflict resolution preserves the 4.1 `void` writer interface and existing pre-rowset aggregation behavior. All upstream large-bitmap tests are included unchanged. There are no configuration, protocol, or persistent-format changes. ### Testing - ASAN Cloud UT compiled with `BUILD_TYPE=ASAN ENABLE_PCH=OFF ./run-cloud-ut.sh -j100`; the packaged repository runner executed `meta_service_test` with `*DeleteBitmap*:*PreRowset*` against an isolated FoundationDB 7.3.69 instance. - **56 passed, 1 skipped**, with zero failures/errors. All **16 real FDB cases passed**. MemTxnKv passed 15 new cases and skipped the range-clear conflict case that its conflict detector cannot model. Coverage includes large v1/v2 and dual writes, compaction/schema change, overlapping cleanup/replacement, ambiguous commits, lock loss, shorter retries, explicit sub-transactions, and unlocked atomic overwrites. - The test runtime used the bundled FDB 7.3.69 external-client directory; no ASAN failure was suppressed. - `ENABLE_PCH=OFF ENABLE_INJECTION_POINT=OFF ./build.sh --cloud -j100` passed for the production Release build with `MAKE_TEST=OFF`. - clang-format 16 and `git diff --check upstream/branch-4.1...HEAD` passed. FE/BE end-to-end tests were not run. As in #70478, the pre-existing race involving delayed cleanup after an RPC has returned OK remains outside this change. ### Release note Fix Cloud Merge-on-Write writes failing when a single delete bitmap exceeds the FoundationDB transaction limit. ### Check List (For Author) - [x] Unit Test - [x] Behavior changed: an oversized locked bitmap can span metadata transactions while retaining the existing publish boundary. - [x] No documentation change required. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
