Smartling <-> Localize fixups

[stevejalim]

Sent an initial batch of fivev pages into Smartling to be translated into French.

Smartling sync failed on two pages

• Fast VPN from Mozilla
• What is an IP Address

It's unclear which Page caused which error, but the first error seen (which I think was for What is an IP Address?) was:

"wagtail_localize_smartling.api.client.InvalidResponse: Response did not match expected format: {'response': {'code': 'VALIDATION_ERROR', 'errors': [{'key': None, 'message': 'File locked: job_7.po', 'details': None}]}}"

• Draftsharing refused to provide a HTML dump of a linked page - possibly because the link had expired (and possibly isn't being correctly invalidated)

We need some tightening up

Smartling validation error

Preview Give feedback

1. Replicate Smartling validation error locally, using data from prod CANNOT REPLICATE LOCALLY 😡
2. Work around issue
3. Ensure translations flow back as expected

wagtaildraftsharing fixups

Preview Give feedback

1. Trace through how a Smartling Job might have picked up an expired link
2. Allow creating multiple links per Revision, so we can support different max_ttls
3. New release and integration into bedrock

Wagtail-localize-smartling improvements

Preview Give feedback

1. "File locked" failure during upload wagtail-localize-smartling#45
   1 of 3
   +0
   
2. More logging on the w-l-s side, with better breadcrumbs - surface Smartling error codes in InvalidResponse
3. Surface the smartling HTTP error code in the custom InvalidResponse exception
4. Handle failed job creation better - don't leave things in a state were next time we can't retry
5. "Empty job problem" when resubmitting a page that has had a translation deleted

Bedrock changes

Preview Give feedback

1. Generate a sharing link with no max TTL
2. Add Sentry logging to visual context callback
3. Worst case in our visual context callback at least raise IncapableVisualContextCallback to stop things failing
4. Consider pushing the translation job into the async queue, rather than wait for the cron?

[stevejalim]

So using an export of the CMS data from prod, I couldn't replicate the initial sync problem with either of the pages that seemed problematic.

What we can do for now, though, is make things more robust, and deal with the issue around the expired draftsharing link