storage: add gc benchmark with multiple range tombstones #87417
Assignees
Labels
A-kv-replication
Relating to Raft, consensus, and coordination.
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
Comments
aliher1911
added
C-enhancement
|
cc @cockroachdb/replication |
|
Bench results for ref: |
craig bot
pushed a commit
that referenced
this issue
Sep 26, 2022
88252: storage: add gc benchmarks with multiple range tombstones r=erikgrinaker a=aliher1911 Previously we only had tests that verify GC performance with no range tombstones or with a single range tombstone under all the data. This commit extends the benchmark to have many tombstones interleaving data at different keys. Release note: None Fixes #87417 88295: sql: add assertion for txn passed to internal executor r=ajwerner a=ZhouXing19 If an internal executor is created with a txn binding to it, it should not be used to execute a statement with a nil txn. Release note: None 88670: schemachanger/rel: fix race due to failure to clone constraint slots r=ajwerner a=ajwerner The fundamental race here is that while the slots themselves were being copied by value, the "any" clauses which are a slice were not. The second bug here is that the "inline" values were not being properly reset. That bug could lead to problems when the query was run again in the context of a different element set. We need to reset those inline values too. Fixes #88628 Release note: None 88716: roachprod: set default cluster settings when starting r=yuzefovich,srosenberg,smg260 a=renatolabs In #88514, the cluster start logic was refactored to reuse the same code across `init` and `start`, fixing a bug in the former. However, the refactoring overlooked the fact that we previously always set the default cluster settings when there's more than one node in the cluster. This fixes that by setting the default cluster settings in that case; one particularly important cluster setting is the license key, necessary for some roachtests. Fixes #88660 Fixes #88665 Fixes #88666 Fixes #88710. Release note: None Co-authored-by: Oleg Afanasyev <oleg@cockroachlabs.com> Co-authored-by: Jane Xing <zhouxing@uchicago.edu> Co-authored-by: Andrew Werner <awerner32@gmail.com> Co-authored-by: Renato Costa <renato@cockroachlabs.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Currently we only have a bench which adds a single range tombstone to minimally measure regression in performance.
We need a better coverage if we want to enable mvcc range keys for everyone.
Benchmark should add N range tombstones and make only some of them cover measured GC requests. For example N keys should be spread across the full range of versions, not only bottom N point keys as that would be non-representative.
Jira issue: CRDB-19336
Epic CRDB-2624
The text was updated successfully, but these errors were encountered: