Axios timeouts are not properly handled

[benoit74]

See https://axios-http.com/docs/cancellation

From what I read there and what I find in the code, it seems that:

• Axios by itself only handles timeouts on response-related timeout
• Axios does not handles other timeouts linked to connection issues, it will only timeout when caller timeout
• mwoffliner has no logic to handle these connection issues
• mwoffliner been a "client" application, it has no caller/stack timeouts like a webserver would have, so in case of connection issues it never timeouts

I consider this is a very probable explanation of some issues we've seen around mwoffliner been like frozen / dead

[kelson42]

@benoit74 Seems a promising way of investigation indeed! AFAIK we are stuck we a relatively old version of Axios.

[Optimus-NP]

We currently use the Axios API in 9 places across our codebase. Below is a summary of its usage:

File	Method	Has Catch Block	Has Timeout	Uses validateStatus
util/dump.ts	GET	No	Yes (60000)	Yes
util/misc.ts	GET	No	No	No
util/mw-api.ts	GET	No	No	No
sanitize-argument.ts	GET	Yes	No	No
sanitize-argument.ts	GET	Yes	No	No
MediaWiki.ts	POST	No	No	No
Downloader.ts	GET	Yes	No	No
Downloader.ts	GET	No	No	No
Downloader.ts	GET	No	No	No
mwoffliner.lib.ts	GET	Yes	No	No

To improve reliability and consistency, I plan to implement an Axios wrapper that will:

• Enforce a timeout for all requests.
• Support early aborting if an HTTP connection cannot be established.
• Implement automatic retries for requests that fail due to timeouts.
• Include validation status checks to prevent retries on 4XX errors.

This wrapper will help standardize API calls across our codebase and improve fault tolerance.

[benoit74]

Issue is already assigned, I'm already working on it and a PR is even under way (not yet totally ready, but close to).