[
https://issues.apache.org/jira/browse/TIKA-4813?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Allison updated TIKA-4813:
------------------------------
Description:
Problem: Tika's timeout system was three incoherent knobs. A parser's own
configured timeout got overwritten by an unrelated stall-detector setting.
Timeouts couldn't compose across nesting (PDF → zip → OCR). A single timed-out
subprocess could kill the whole worker JVM instead of just failing that one
document.
Proposal: every external call gets min(what it asked for, time left on the
task). ParseTimeout (new, in tika-core) holds that remaining-time clock per
task, shared automatically across recursion. ProcessUtils/TikaHttpClient apply
the rule internally and checkpoint the clock while waiting, so a legitimately
slow call doesn't look like a hang. TikaTimeoutException is now catchable (was
an uncaught RuntimeException), so one slow embedded file gets recorded and
skipped instead of aborting the whole document; running out of total time now
truncates cleanly instead of ending the pipes process.
was:
Problem: Tika's timeout system was three incoherent knobs. A parser's own
configured timeout got overwritten by an unrelated stall-detector setting.
Timeouts couldn't compose across nesting (PDF → zip → OCR). A single timed-out
subprocess could kill the whole worker JVM instead of just failing that one
document.
Proposal: every external call gets min(what it asked for, time left on the
task). ParseTimeout (new, in tika-core) holds that remaining-time clock per
task, shared automatically across
recursion. ProcessUtils/TikaHttpClient apply the rule internally and
checkpoint the clock while waiting, so a legitimately slow call doesn't look
like a hang. TikaTimeoutException is now catchable (was an uncaught
RuntimeException), so one slow embedded file gets recorded and skipped instead
of aborting the whole document; running out of total time now truncates cleanly
instead of crashing the JVM.
> Clean up timeout logic in 4.x
> -----------------------------
>
> Key: TIKA-4813
> URL: https://issues.apache.org/jira/browse/TIKA-4813
> Project: Tika
> Issue Type: Task
> Reporter: Tim Allison
> Priority: Minor
>
> Problem: Tika's timeout system was three incoherent knobs. A parser's own
> configured timeout got overwritten by an unrelated stall-detector setting.
> Timeouts couldn't compose across nesting (PDF → zip → OCR). A single
> timed-out subprocess could kill the whole worker JVM instead of just failing
> that one document.
> Proposal: every external call gets min(what it asked for, time left on the
> task). ParseTimeout (new, in tika-core) holds that remaining-time clock per
> task, shared automatically across recursion. ProcessUtils/TikaHttpClient
> apply the rule internally and checkpoint the clock while waiting, so a
> legitimately slow call doesn't look like a hang. TikaTimeoutException is now
> catchable (was an uncaught RuntimeException), so one slow embedded file gets
> recorded and skipped instead of aborting the whole document; running out of
> total time now truncates cleanly instead of ending the pipes process.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
