noticeably slower than core regexp

[stevenh]

Gave this a go with our patterns which are processing a huge file of log entries and it's unfortunately significantly slower.

It could be I need a config but thought you would be interested in the example.

The code uses these regexps:

		idPattern:      regexp.MustCompile(`^(\d+|[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}|[0-9a-fA-F]{32})$`),
		phpPattern:     regexp.MustCompile(`^/.*[\w-]+\.php`),

It uses a simple:

idPattern.MatchString(path)
phpPattern.MatchString(path)

With core regexp it takes ~1m21s to process 10m lines, with coregex it takes 1m35s.

[kolkov]

Thank you very much! The problem has been localized and fixed.

[stevenh]

I tried this on our latest version, which has a slightly changed pattern but its still quite a bit slower than core regexp.

These are the patterns in use now:

		id:           regexp.MustCompile(`^(\d+|[0-9a-fA-F]{8}-?[0-9a-fA-F]{4}-?[0-9a-fA-F]{4}-?[0-9a-fA-F]{4}-?[0-9a-fA-F]{12})$`),
		php:          regexp.MustCompile(`^/.*[\w-]+\.php`),

With core regexp we get 151327 rows/s but that drops to 132629 rows/s with coregx, which is up from the last version which was 127803 rows/s.

The code is reading lines from a file, parsing json then performing some check with the two patterns above.

The test machine is gce n2-standard-4 and the test is running with 8 workers processing lines from single producer.

I tried both with worker localised regexp and common, no change.

[stevenh]

The above results were using 0.10.6, updating to 0.10.7 shows ~ 30% slow down dropping from 132629 rows/s to 100931 rows/s

[stevenh]

This is the cpu profile with 0.10.7 with worker local regexp:

with 0.10.6

and stdlb for reference: