Describe the problem
Currently, on the 22.2 release branch there is a sizable performance regression in EncodeMVCCKey, relative to 22.1:
$ benchstat bench-22.1 bench-22.2
name old time/op new time/op delta
EncodeMVCCKey/key=empty/ts=empty-16 9.85ns ± 3% 14.73ns ± 1% +49.49% (p=0.000 n=9+7)
EncodeMVCCKey/key=empty/ts=walltime-16 11.7ns ± 2% 12.3ns ± 4% +5.10% (p=0.000 n=10+10)
EncodeMVCCKey/key=empty/ts=walltime+logical-16 12.3ns ± 3% 12.6ns ± 4% ~ (p=0.072 n=10+10)
EncodeMVCCKey/key=empty/ts=all-16 12.5ns ± 2% 12.7ns ± 4% ~ (p=0.315 n=8+10)
EncodeMVCCKey/key=short/ts=all-16 13.0ns ± 2% 12.8ns ± 4% ~ (p=0.247 n=10+10)
EncodeMVCCKey/key=short/ts=empty-16 10.3ns ± 4% 15.0ns ± 3% +45.29% (p=0.000 n=10+10)
EncodeMVCCKey/key=short/ts=walltime-16 11.9ns ± 1% 12.7ns ± 4% +6.61% (p=0.000 n=9+10)
EncodeMVCCKey/key=short/ts=walltime+logical-16 12.7ns ± 4% 12.8ns ± 4% ~ (p=0.579 n=10+10)
EncodeMVCCKey/key=long/ts=all-16 36.7ns ± 2% 37.8ns ± 4% +3.08% (p=0.013 n=9+10)
EncodeMVCCKey/key=long/ts=empty-16 36.1ns ± 4% 39.2ns ± 2% +8.63% (p=0.000 n=10+10)
EncodeMVCCKey/key=long/ts=walltime-16 37.1ns ± 4% 38.0ns ± 2% +2.41% (p=0.015 n=10+10)
EncodeMVCCKey/key=long/ts=walltime+logical-16 37.8ns ± 3% 37.8ns ± 3% ~ (p=0.353 n=10+10)
The above was generated by comparing benchmark results between 22.1 (9eb4da2) to 22.2 (054f32b), via the following:
$ ./dev bench ./pkg/storage --filter EncodeMVCCKey --count 10 -v --stream-output
Jira issue: CRDB-19971
Epic CRDB-2624
Describe the problem
Currently, on the 22.2 release branch there is a sizable performance regression in
EncodeMVCCKey, relative to 22.1:The above was generated by comparing benchmark results between 22.1 (9eb4da2) to 22.2 (054f32b), via the following:
Jira issue: CRDB-19971
Epic CRDB-2624