Failed to classify images - Array buffer allocation failed

[jahknem]

When trying to run the classify process manually, it fails:

root@raspberrypi:/var/www/nextcloud# sudo -u www-data php occ recognize:classify -vvv
Classifying photos of user NextcloudAdmin
Failed to classify images
Classifier process error

I would like to have it run automatically and manually. How do I achieve this?

In the logs I find two log entries which seem to be related: a warning from recognize:

[recognize] Warning: Classifier process output: RangeError: Array buffer allocation failed
    at new ArrayBuffer (<anonymous>)
    at Object.toArrayBuffer (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/io_utils.js:41:18)
    at NodeFileSystem.<anonymous> (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:268:76)
    at step (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:48:23)
    at Object.next (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:29:53)
    at fulfilled (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:20:58)


at 2021-09-16T14:16:03+02:00

and an error from php. Sorry that it is not formatted, I could not find a formatted version.

{"reqId":"T76flJzxdo72sKZGbL2x","level":3,"time":"2021-09-16T14:16:03+02:00","remoteAddr":"","user":"--","app":"PHP","method":"","url":"--","message":{"Exception":"Error","Message":"fwrite(): write of 8192 bytes failed with errno=32 Broken pipe at /var/www/nextcloud/3rdparty/symfony/process/Pipes/AbstractPipes.php#128","Code":0,"Trace":[{"function":"onError","class":"OC\\Log\\ErrorHandler","type":"::"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Pipes/AbstractPipes.php","line":128,"function":"fwrite"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Pipes/UnixPipes.php","line":95,"function":"write","class":"Symfony\\Component\\Process\\Pipes\\AbstractPipes","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1436,"function":"readAndWrite","class":"Symfony\\Component\\Process\\Pipes\\UnixPipes","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1359,"function":"readPipes","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1407,"function":"updateStatus","class":"Symfony\\Component\\Process\\Process","type":"->","args":["*** sensitive parameters replaced ***"]},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":655,"function":"readPipesForOutput","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/apps/recognize/lib/Service/ClassifyImagenetService.php","line":82,"function":"getIterator","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/apps/recognize/lib/Command/Classify.php","line":99,"function":"classify","class":"OCA\\Recognize\\Service\\ClassifyImagenetService","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Command/Command.php","line":255,"function":"execute","class":"OCA\\Recognize\\Command\\Classify","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":1000,"function":"run","class":"Symfony\\Component\\Console\\Command\\Command","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":271,"function":"doRunCommand","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":147,"function":"doRun","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/lib/private/Console/Application.php","line":215,"function":"run","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/console.php","line":100,"function":"run","class":"OC\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/occ","line":11,"args":["/var/www/nextcloud/console.php"],"function":"require_once"}],"File":"/var/www/nextcloud/lib/private/Log/ErrorHandler.php","Line":91,"CustomMessage":"--"},"userAgent":"--","version":"20.0.10.1","id":"614335b7e054c"}

I am running Nextcloud 20.0.10 on a raspberry pi 4 with 4 GB RAM. More than 3 GB RAM are available

Operating System: | Linux 5.10.17-v7l+ armv7l CPU: | ARMv7 Processor rev 3 (v7l) (4 cores) Memory: | 3.74 GB

[arch-user-france1]

I get an error like 'classifier timed out' But only appears when the process begins running and then gets terminated @marcelklehr please make the timeout higher. My computer isn't high-end and just may need a bit longer. Classifier works but most times a timed out returns. I don't know if this also affects automatic launches

[arch-user-france1]

When trying to run the classify process manually, it fails:

root@raspberrypi:/var/www/nextcloud# sudo -u www-data php occ recognize:classify -vvv
Classifying photos of user NextcloudAdmin
Failed to classify images
Classifier process error

I would like to have it run automatically and manually. How do I achieve this?

In the logs I find two log entries which seem to be related: a warning from recognize:

[recognize] Warning: Classifier process output: RangeError: Array buffer allocation failed
    at new ArrayBuffer (<anonymous>)
    at Object.toArrayBuffer (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/io_utils.js:41:18)
    at NodeFileSystem.<anonymous> (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:268:76)
    at step (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:48:23)
    at Object.next (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:29:53)
    at fulfilled (/var/www/nextcloud/apps/recognize/node_modules/@tensorflow/tfjs-node/dist/io/file_system.js:20:58)


at 2021-09-16T14:16:03+02:00

and an error from php. Sorry that it is not formatted, I could not find a formatted version.

{"reqId":"T76flJzxdo72sKZGbL2x","level":3,"time":"2021-09-16T14:16:03+02:00","remoteAddr":"","user":"--","app":"PHP","method":"","url":"--","message":{"Exception":"Error","Message":"fwrite(): write of 8192 bytes failed with errno=32 Broken pipe at /var/www/nextcloud/3rdparty/symfony/process/Pipes/AbstractPipes.php#128","Code":0,"Trace":[{"function":"onError","class":"OC\\Log\\ErrorHandler","type":"::"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Pipes/AbstractPipes.php","line":128,"function":"fwrite"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Pipes/UnixPipes.php","line":95,"function":"write","class":"Symfony\\Component\\Process\\Pipes\\AbstractPipes","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1436,"function":"readAndWrite","class":"Symfony\\Component\\Process\\Pipes\\UnixPipes","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1359,"function":"readPipes","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":1407,"function":"updateStatus","class":"Symfony\\Component\\Process\\Process","type":"->","args":["*** sensitive parameters replaced ***"]},{"file":"/var/www/nextcloud/3rdparty/symfony/process/Process.php","line":655,"function":"readPipesForOutput","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/apps/recognize/lib/Service/ClassifyImagenetService.php","line":82,"function":"getIterator","class":"Symfony\\Component\\Process\\Process","type":"->"},{"file":"/var/www/nextcloud/apps/recognize/lib/Command/Classify.php","line":99,"function":"classify","class":"OCA\\Recognize\\Service\\ClassifyImagenetService","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Command/Command.php","line":255,"function":"execute","class":"OCA\\Recognize\\Command\\Classify","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":1000,"function":"run","class":"Symfony\\Component\\Console\\Command\\Command","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":271,"function":"doRunCommand","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/3rdparty/symfony/console/Application.php","line":147,"function":"doRun","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/lib/private/Console/Application.php","line":215,"function":"run","class":"Symfony\\Component\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/console.php","line":100,"function":"run","class":"OC\\Console\\Application","type":"->"},{"file":"/var/www/nextcloud/occ","line":11,"args":["/var/www/nextcloud/console.php"],"function":"require_once"}],"File":"/var/www/nextcloud/lib/private/Log/ErrorHandler.php","Line":91,"CustomMessage":"--"},"userAgent":"--","version":"20.0.10.1","id":"614335b7e054c"}

I am running Nextcloud 20.0.10 on a raspberry pi 4 with 4 GB RAM. More than 3 GB RAM are available

Operating System: | Linux 5.10.17-v7l+ armv7l CPU: | ARMv7 Processor rev 3 (v7l) (4 cores) Memory: | 3.74 GB

I think that the raspberry pi kernel doesn't even support fsync - and fsync command fails so you have to try with a patch or another kernel that works with the pi

Edits It more looks like a process is terminated (pipe broken) And this might be because of low specs...

Do you have a swap (please do not set up a swap on a SD card it will die

[marcelklehr]

It does seem like your RAM is exhausted when running the classifier. Can you confirm that by monitoring running processes in e.g. htop?

[jahknem]

When starting htop with -d 1 I can see RAM Usage increases to 1.5-1.6 GB before the process dies.

After I increased the swap it took a lot longer for the process to die but it still did with the same error. It also never reached more than 50% RAM Usage.

Is there a way I can get more verbose debug information?

[arch-user-france1]

Do you have earlyoom or something like that? Earlyoom is a fix of the oom-killer

But still, the percent when it kills a process is too high. It might cause troubles

Also maybe something's not good with the ram? I heard that a RAM of a PI4 can be locked so you can't allocate all. Simply unlock by an SD card with an image from the raspberry pi flasher that will change this setting

[arch-user-france1]

It looks like something's not good with fsync. An fwrite command fails with only 8MB writing. Do you have free space on the raspberry pi? Or isn't it supported?

I am not a specialist but ... You could Google for it. Fsync Raspberry pi broken pipe etc.

[marcelklehr]

The borken pipe is likely because the subprocess dies while the php process still wants to write to it.

[arch-user-france1]

Maybe an issue with the kernel, couldn't it be?

[jahknem]

Do you have earlyoom or something like that? Earlyoom is a fix of the oom-killer

But still, the percent when it kills a process is too high. It might cause troubles

Also maybe something's not good with the ram? I heard that a RAM of a PI4 can be locked so you can't allocate all. Simply unlock by an SD card with an image from the raspberry pi flasher that will change this setting

I haven't installed earlyoom or anything like that. Tbh I did not even know oom existed. Currently running memtester and I seem to be able to use all of the memory just fine. So that should not be it.

Maybe an issue with the kernel, couldn't it be?

I will try to update and report back

[arch-user-france1]

An update might not help much but try it... Almost always good to do that

Maybe there's another kernel for Raspberry Pi's Why not try then

[jahknem]

The update did not help sadly. The Kernel was also updated. journalctl does not show any kernel errors, so I doubt the kernel is at fault..

Is it possible to have this plugin run on a different system than the nextcloud instance? While using the files on the nextcloud instance?

[marcelklehr]

It may be the architecture. ARM64 does not seem to be supported by libtensorflow.

[jahknem]

It may be the architecture. ARM64 does not seem to be supported by libtensorflow.

I exchanged libtensorflow against a version explicitely compiled for the raspberry as described here: https://qengineering.eu/install-tensorflow-2.1.0-on-raspberry-pi-4.html However I sadly still have the same problem.

[arch-user-france1]

I think at this point I can't help anymore. I'm not contributing code - I didn't look into the code

Are there any Raspberry Pi Simulators here? I don't have version 4 around

Maybe you changed a knob (kernel behaviour setting)?

[metbosch]

Could it be related to the php memory limit? I'm having a similar issue in a rasspi 4, and the top process before the crash was a php one.

[arch-user-france1]

I have it limited to 512MB wich ist recommended by Nextcloud

Could you try to run the script manually without Nextcloud?

[arch-user-france1]

The update did not help sadly. The Kernel was also updated. journalctl does not show any kernel errors, so I doubt the kernel is at fault..

Is it possible to have this plugin run on a different system than the nextcloud instance? While using the files on the nextcloud instance?

Well, I made a request (see Issue somewhere there opened) wich means that it could take some time for that but currently it's not out

[igor-cali]

Same issue for me on Raspbian.

Operating System: Linux 5.10.63-v7l+ armv7l CPU: ARMv7 Processor rev 3 (v7l) (4 cores) Memory: 3.74 GB

[marcelklehr]

Does it work in pure JS mode at least or does that also not work?

[morta5]

For me, JavaScript mode doesn't work either (I am also running Nextxloud 22 on a raspberry pi 4 with yunohost)

[igor-cali]

JS mode does not work for me either.

[arch-user-france1]

Could it be related to the php memory limit? I'm having a similar issue in a rasspi 4, and the top process before the crash was a php one.

No, it's not related to it because it runs using node not php But the thing that searches for new filed is using php

[marcelklehr]

Closing due to being outdated. v2 is no longer maintained. Please try again with Recognize v3 and Nextcloud 25 :pray: