go.mod: revert comparer change and bump Pebble to 0f785fec58c0

[RaduBerinde]

The comparer changes effectively revert #128043, which can cause
replica inconsistency during/after upgrades.

Changes:

• 0f785fec metamorphic: abridge failure output
• be56747f db: fix overlap check for flushable ingest excises
• 2569414a db: remove race in TestCrashOpenCrashAfterWALCreation
• 575f7a04 sstable: support columnar blocks in Layout.Describe
• c88c7471 github: fix code cover publish workflow
• c0fa4a9c sstable: populate CompareSuffixes on test4bSuffixComparer
• 3a76074f sstable: set IndexPartitions property in columnar sstable writer
• 0595c1fb colblk: define behavior of KeyWriter.ComparePrev with no previous
• 01dcf575 base: make comparer tolerate empty keys
• d73ab80f db: allow excises to unconditionally be flushable ingests
• b34a3937 base: allow CompareSuffixes to be stricter than Compare
• 90356021 db: refactor replayWAL to use flushes to make versionEdits

Informs: #130533

Release note: none.
Epic: none.

[cockroach-teamcity]

This change is 

[jbowens]

Reviewed 8 of 8 files at r1, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @itsbilal)

[itsbilal]

Might be worth holding off on this for at least a day so we can get the fix for cockroachdb/pebble#3963

[RaduBerinde]

Yes, agreed.

[RaduBerinde]

Updated to the current Pebble master.

[itsbilal]

LGTM

[RaduBerinde]

I'm getting this test error:

=== RUN   TestStoreMetrics
    test_log_scope.go:170: test logs captured to: outputs.zip/logTestStoreMetrics1690765082
    test_log_scope.go:81: use -show-logs to present logs inline
    client_metrics_test.go:389: counter rocksdb.block.cache.hits = 0 < min 10
    client_metrics_test.go:389: counter rocksdb.block.cache.hits = 0 < min 10
    client_metrics_test.go:393: -- test log scope end --

Any ideas why this changed?

[jbowens]

I don't like that test. I've run afoul of it before, because it ends up asserting on details of the storage engine mechanics. I've got no idea why we'd be getting zero block cache hits though.

[itsbilal]

Looking at the logs, it's sometimes possible for the keys to just stick around in the WAL/memtable. Let me see if I can force a flush in the test.

[itsbilal]

This should deflake this test:

deflake.patch

diff --git a/pkg/kv/kvserver/client_metrics_test.go b/pkg/kv/kvserver/client_metrics_test.go
index 232f9c9cfbf..eb6413c6ce9 100644
--- a/pkg/kv/kvserver/client_metrics_test.go
+++ b/pkg/kv/kvserver/client_metrics_test.go
@@ -262,7 +262,7 @@ func TestStoreMetrics(t *testing.T) {
                        },
                }
                stickyServerArgs[i] = base.TestServerArgs{
-                       CacheSize:  1 << 20, /* 1 MiB */
+                       CacheSize:  2 << 20, /* 2 MiB */
                        StoreSpecs: []base.StoreSpec{spec},
                        Knobs: base.TestingKnobs{
                                Server: &server.TestingKnobs{

[RaduBerinde]

Thank you!! What's the explanation why a 1MB cache leads to no hits?

[RaduBerinde]

bors r+

[craig]

Build succeeded:

• docker_image_amd64
• lint
• windows_build
• examples_orms
• local_roachtest
• linux_arm64_build
• linux_amd64_build
• macos_amd64_build
• acceptance
• linux_amd64_fips_build
• unit_tests
• check_generated_code
• macos_arm64_build
• local_roachtest_fips

[itsbilal]

Thank you!! What's the explanation why a 1MB cache leads to no hits?

I actually don't know either, but I also wasn't able to really understand what the test was doing so....