Commit 30bfa2a4 authored by Yanqin Jin's avatar Yanqin Jin Committed by Andrew Kryczka
Browse files

Report corrupted keys during compaction (#7124)

Currently, RocksDB lets compaction to go through even in case of
corrupted keys, the number of which is reported in CompactionJobStats.
However, RocksDB does not check this value. We should let compaction run
in a stricter mode.

Temporarily disable two tests that allow corrupted keys in compaction.
With this PR, the two tests will assert(false) and terminate. Still need
to investigate what is the recommended google-test way of doing it.
Death test (EXPECT_DEATH) in gtest has warnings now.

Pull Request resolved:

Test Plan: make check

Reviewed By: ajkr

Differential Revision: D22530722

Pulled By: riversand963

fbshipit-source-id: 6a5a6a992028c6d4f92cb74693c92db462ae4ad6
parent 3e1eb99a
# Rocksdb Change Log
## Unreleased
### Bug Fixes
* Make compaction report InternalKey corruption while iterating over the input.
## 6.11.3 (7/9/2020)
### Bug Fixes
* Fix a bug when index_type == kTwoLevelIndexSearch in PartitionedIndexBuilder to update FlushPolicy to point to internal key partitioner when it changes from user-key mode to internal-key mode in index partition.
......@@ -246,6 +246,7 @@ void CompactionIterator::NextFromInput() {
if (!ParseInternalKey(key_, &ikey_)) {
// If `expect_valid_internal_key_` is false, return the corrupted key
// and let the caller decide what to do with it.
// TODO(noetzli): We should have a more elegant solution for this.
......@@ -258,7 +259,6 @@ void CompactionIterator::NextFromInput() {
has_current_user_key_ = false;
current_user_key_sequence_ = kMaxSequenceNumber;
current_user_key_snapshot_ = 0;
valid_ = true;
......@@ -895,9 +895,10 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
sub_compact->c_iter.reset(new CompactionIterator(
input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
&existing_snapshots_, earliest_write_conflict_snapshot_,
snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_), false,
&range_del_agg, sub_compact->compaction, compaction_filter,
shutting_down_, preserve_deletes_seqnum_, manual_compaction_paused_,
snapshot_checker_, env_, ShouldReportDetailedTime(env_, stats_),
/*expect_valid_internal_key=*/true, &range_del_agg,
sub_compact->compaction, compaction_filter, shutting_down_,
preserve_deletes_seqnum_, manual_compaction_paused_,
auto c_iter = sub_compact->c_iter.get();
......@@ -395,7 +395,7 @@ TEST_F(CompactionJobTest, Simple) {
RunCompaction({ files }, expected_results);
TEST_F(CompactionJobTest, SimpleCorrupted) {
TEST_F(CompactionJobTest, DISABLED_SimpleCorrupted) {
auto expected_results = CreateTwoFiles(true);
......@@ -989,7 +989,7 @@ TEST_F(CompactionJobTest, MultiSingleDelete) {
// single deletion and the (single) deletion gets removed while the corrupt key
// gets written out. TODO(noetzli): We probably want a better way to treat
// corrupt keys.
TEST_F(CompactionJobTest, CorruptionAfterDeletion) {
TEST_F(CompactionJobTest, DISABLED_CorruptionAfterDeletion) {
auto file1 =
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment