[ https://issues.apache.org/jira/browse/TIKA-3103?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17112635#comment-17112635 ]
Tim Allison edited comment on TIKA-3103 at 5/20/20, 9:54 PM: ------------------------------------------------------------- >Maybe it's something to do with concurrency in the Tika server instead? I test it regularly with around 8 concurrent requests. That shouldn't be a problem. I don't often work with tesseract, tho... > I'm not sure. Where do I find out? Both the child and parent should log when there's a restart. I agree about expected behavior. From an implementation standpoint, though, if the child process is killed, even if we add a shutdown hook to kill tesseract, there's no guarantee that it will be executed. If you can share the logs and/or if we can identify that the restarts are orphaning the children and there's not some other way for the process to be orphaned, that would help. was (Author: talli...@mitre.org): >Maybe it's something to do with concurrency in the Tika server instead? I test it regularly with around 8 concurrent requests. That shouldn't be a problem. I don't often work with tesseract, tho... > I'm not sure. Where do I find out? Both the child and parent should log when there's a restart. I agree about expected behavior. From an implementation standpoint, though, if the child process is killed, even if we add a shutdown hook to kill tesseract, there's no guarantee that it will be executed. > Tesseract fails to respect timeouts and clean up after itself > ------------------------------------------------------------- > > Key: TIKA-3103 > URL: https://issues.apache.org/jira/browse/TIKA-3103 > Project: Tika > Issue Type: Bug > Components: ocr > Affects Versions: 1.24.1 > Reporter: Radim Rehurek > Priority: Critical > > We're using the Tika Server with OCR: > _java -jar /opt/tika/tika-server-1.24.1.jar -p 9998 -spawnChild -JXmx500m_ > > Two undersirable things happen: > h3. 1. The CPU runs at 100% for >10 minutes, long after any Tika requests > have finished. > These processes show in _top_ as "tesseract" (OCR) and consume all CPU cores > at 100%. > They eventually die (or finish?) but the machine is unusable in the mean time. > *Expected behaviour:* Tika cleans up spawned processes after itself: at most > after its timeout limit (which is 2 minutes I believe?) > h3. 2. The temp is full of files like: > {{root@38acd588ee22:/# ll /tmp/}} > {{total 197320}} > {{drwxrwxrwt 1 root root 24576 May 20 09:35 ./}} > {{drwxr-xr-x 1 root root 4096 May 20 08:40 ../}} > {{-rw-r--r-- 1 root root 9273920 May 20 08:56 > TIKA_streamstore_11144988934311367241.tmp}} > {{-rw-r--r-- 1 root root 8938048 May 20 08:57 > TIKA_streamstore_11649337406504198407.tmp}} > {{-rw-r--r-- 1 root root 9478720 May 20 08:56 > TIKA_streamstore_13551529918743702933.tmp}} > {{-rw-r--r-- 1 root root 9151040 May 20 08:57 > TIKA_streamstore_13568226047805501311.tmp}} > {{-rw-r--r-- 1 root root 7701056 May 20 08:56 > TIKA_streamstore_13908373602714189455.tmp}} > {{…}} > {{-rw-r--r-- 1 root root 33367 May 20 08:55 > apache-tika-11167866320029165062.tmp}} > {{-rw-r--r-- 1 root root 44353 May 20 08:54 > apache-tika-1152515137515755865.tmp}} > {{-rw-r--r-- 1 root root 245279 May 20 08:52 > apache-tika-12106368488659105236.tmp}} > {{-rw-r--r-- 1 root root 1759 May 20 08:47 > apache-tika-12291680472524021463.tmp}} > {{…}} > > slowly filling up the disk. > *Expected behaviour*: Tika cleans up disk space after itself. > > These bugs are critical for us. What's the best way to avoid them? > -- This message was sent by Atlassian Jira (v8.3.4#803005)