Update jquery.couch.js

[andrewspode]

In the original code, if there was an error doing the long poll, it would wait 100ms and then try again. If it errors again, it doubles to 200ms. If it errors again, it doubles again etc. with no cap.

This means after 10 errors in a row, you will be in a position of waiting a minute and a half before it attempts to reconnect. After that, 3 minutes and then it frankly spirals out of control!

That means if you lose internet connectivity (entirely possible if your wireless/3g is flakey) - you will very quickly, essentially lose your polling abilities.

My suggestion is to put a cap on this timeout - I've simply added in a conditional so that it will never go above 5 seconds. In my own use case, I'm actually going to keep it at 100ms and not multiply, so it might be worth having this as an option instead of hard coded.

[tilgovi]

I think 5s is too low. We can be much smarter about this whole thing, IMO.

What do you think of

• Backoff 2x as before
• Limit the backoff only to cases where the connection timed out. Losing connectivity should mean total, fast failure in many cases, and we could detect the difference between these cases. In the case of server error or no connectivity, that error should bubble immediately and the application can decide what to do (either sleep, back off, or hook into the connectivity notification APIs that are cropping up in modern browsers)
• Add a random jitter to the back-off. When timeouts are happening because the server is overwhelmed it could help to prevent clients from stampeding. Such a jitter is a classic feature of backoffs like this.

Finally, I haven't looked at this code in a while, but we should be sure that the whole thing is cancellable by the application in case the backoff gets large.

[andrewspode]

Agreed. Rather selfishly, for my own uses I just want it to keep trying rather than back off at all. So this is more of a "are you aware of this issue" rather than an actual proposed fix.

I'm not sure who is looking after this plugin - but I'm happy to code these additions if it would be appreciated, otherwise the current maintainer might want to look at the issue.

[nomicode]

The whole community maintains it, so feel free to submit another changeset!

[tilgovi]

@unclespode Cool. Thanks for the pull request. I'd love to see the fuller solution (distinguish timeouts from fast, total connection failure), but it's not critical. If you get the time to improve this please put it here. As is, though, I'm ambivalent about what is the better default.

[andrewspode]

I've been working on the code as I develop my project at the moment. I've actually written EventSource (SSE) support into the changes function, so that it prefers this over longpolling, with a fallback to long polling where it isn't available, with no discernible difference as far as the coder is concerned.

I won't submit the patch until it's finalised.

[wohali]

jquery.couch.js has been moved to https://github.com/apache/couchdb-jquery-couch . If you are still keen on addressing this issue, and the issue still exists, please file a new PR on that repository.