Page MenuHomePhabricator

CX2: Big sections are untranslatable
Closed, ResolvedPublic

Description

Some bigger pieces of content are impossible to translate in CX2.
For example, big tables in en:List_of_English_football_champions, like one in section "Football League First Division (1892–1992)".

We get "Automatic translation failed" messages and server response is following:

{
    "status": 500,
    "type": "internal_error",
    "title": "PayloadTooLargeError",
    "detail": "request entity too large",
    "method": "POST",
    "uri": "/v2/translate/en/es/Apertium"
}

Capturing error on cxserver gives:

{
    "message": "request entity too large",
    "expected": 107790,
    "length": 107790,
    "limit": 102400,
    "type": "entity.too.large"
}

Using "Use source text" does not help either.
Option "Don't use MT" does work, but not really helpful if we need to recreate such a long piece of content.
CX1 handles this case of large sections better.

Details

Related Gerrit Patches:
mediawiki/services/cxserver : masterIncrease the maximum request body size to 0.5mb

Event Timeline

Another report of this problem - T200665#4525554

Restricted Application changed the subtype of this task from "Deadline" to "Task". · View Herald TranscriptAug 23 2018, 9:48 AM
This comment was removed by Etonkovidova.

No. it does not. this is different issue.

Change 456116 had a related patch set uploaded (by Santhosh; owner: Santhosh):
[mediawiki/services/cxserver@master] Increase the maximum request body size to 0.5mb

https://gerrit.wikimedia.org/r/456116

I created https://phabricator.wikimedia.org/T203077 for a general performance analysis of /translate api

Change 456116 merged by Nikerabbit:
[mediawiki/services/cxserver@master] Increase the maximum request body size to 0.5mb

https://gerrit.wikimedia.org/r/456116

Re-checked the same article reported - en:List_of_English_football_champions - the issue is still happening exactly as described in the ticket.

The big tables in sections 'Football League First Division (1892–1992)' and 'Premier League (1992–present)' when being translates will display "Automatic translation failed" message and fail with the 500 internal error.

The sections with smaller tables, e.g. 'Total titles won' or 'Total titles won by region' will be translated without any problem. I am moving the task back to In Progress.

This and other cxserver patches have not been deployed yet.

Moving back to QA, since cxserver is updated now

Stashbot added a subscriber: Stashbot.

Mentioned in SAL (#wikimedia-operations) [2018-09-05T07:23:36Z] <kartik@deploy1001> Started deploy [cxserver/deploy@f341eec]: Update cxserver to 81d1a97 (T202933, T202283, T189438)

Mentioned in SAL (#wikimedia-operations) [2018-09-05T07:27:38Z] <kartik@deploy1001> Finished deploy [cxserver/deploy@f341eec]: Update cxserver to 81d1a97 (T202933, T202283, T189438) (duration: 04m 03s)

@Etonkovidova Can you verify this again or should I do?

Petar.petkovic closed this task as Resolved.Sep 7 2018, 2:40 PM
Petar.petkovic moved this task from QA to Done on the Language-2018-July-September board.
Petar.petkovic removed subscribers: Stashbot, gerritbot.

I have tried again to translate "Football League First Division (1892–1992)" section of en:List_of_English_football_champions to Spanish using Apertium.
After the patch, the section no longer fails with the reported error. Network call to add paragraph with big table took 28 seconds to complete, while table showed 10 seconds later. Saving took 8 seconds.

That error is reported when using Yandex. I have tried to translate the same section from above to Serbian in production, and it still fails. Fallback to source works, but there is no visual clue that fallback is being processed (no spinner, but "Add translation" shows) and fallback took around 25 seconds to complete.