[
https://issues.apache.org/jira/browse/NUTCH-2690?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16815355#comment-16815355
]
Sebastian Nagel commented on NUTCH-2690:
----------------------------------------
PR updated, squashed and rebased to current master.
I'll commit next week, but reviews are welcome. Thanks!
Below the benchmark results from the unit tests. While the new plugin
outperforms urlfilter-regex, the plugin urlfilter-automaton is still faster.
However, the regular expressions supported by the
[dk.brics.automaton](https://www.brics.dk/automaton/) are less expressive, eg.
the "skip URLs with slash-delimited segment that repeats 3+ times" rule cannot
be expressed because there are no back-references.
{noformat}
% ant test
...
% grep 'bench time' build/urlfilter-regex/test/*.txt | sed -E 's@api\..*( -
)@\1... @'
2019-04-11 13:46:18,776 INFO ... - bench time (50) 107ms
2019-04-11 13:46:18,845 INFO ... - bench time (100) 66ms
2019-04-11 13:46:18,961 INFO ... - bench time (200) 116ms
2019-04-11 13:46:19,192 INFO ... - bench time (400) 231ms
2019-04-11 13:46:19,663 INFO ... - bench time (800) 471ms
% grep 'bench time' build/urlfilter-fast/test/*.txt | sed -E 's@api\..*( -
)@\1... @'
2019-04-11 13:48:05,024 INFO ... - bench time (50) 72ms
2019-04-11 13:48:05,112 INFO ... - bench time (100) 84ms
2019-04-11 13:48:05,233 INFO ... - bench time (200) 121ms
2019-04-11 13:48:05,446 INFO ... - bench time (400) 213ms
2019-04-11 13:48:05,687 INFO ... - bench time (800) 241ms
% grep 'bench time' build/urlfilter-automaton/test/*.txt | sed -E 's@api\..*( -
)@\1... @'
2019-04-11 13:43:11,794 INFO ... - bench time (50) 43ms
2019-04-11 13:43:11,834 INFO ... - bench time (100) 37ms
2019-04-11 13:43:11,899 INFO ... - bench time (200) 65ms
2019-04-11 13:43:11,996 INFO ... - bench time (400) 97ms
2019-04-11 13:43:12,175 INFO ... - bench time (800) 178ms
{noformat}
> Configurable and fast URL filter
> --------------------------------
>
> Key: NUTCH-2690
> URL: https://issues.apache.org/jira/browse/NUTCH-2690
> Project: Nutch
> Issue Type: Improvement
> Components: plugin
> Reporter: Sebastian Nagel
> Priority: Minor
> Fix For: 1.16
>
>
> This improvement introduces a new URL filter plugin "urlfilter-fast" (naming
> debatable) which is in use at Common Crawl [since
> 2013|https://github.com/commoncrawl/nutch/commit/968e0d8f292bed46e4e3eb276cb475f4403ea9bd]
> to apply a long list of filters.
> # an exact (suffix) match against the host name is done to retrieve
> host/domain-specific regex rules
> # applies a regular expression against the path (and query) component of the
> URL
> What makes it faster than urlfilter-regex for common cases:
> - regexes are selected by host name or it's domain suffix, so there are
> usually fewer rules to be checked. That's similar to NUTCH-1838 but any
> domain suffix can be matched including {{subdomain.domain.com}}, {{com}} or
> {{.}} for global rules. The selection by host name suffix is considerably
> fast.
> - regexes are applied only to the path component (optionally including the
> query) and not the entire URL.
> Matching against a shorter string can make a huge difference for more
> complex regular expressions.
> - the rule to deny everything from a host or domain gets special treatment to
> be fast
> More details about the rule format are found in the plugin's
> [README|https://github.com/commoncrawl/nutch/blob/cc-fast-url-filter/src/plugin/urlfilter-fast/README.md].
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
