Punchlist

[wlonk]

Braindump:

Immediate things:

• [x] Get parser pipeline stood back up.
• [ ] See if that handles the stream of Celery errors.

Partly investigated:

• [ ] Get more Steam API keys (help scale)
• [ ] Cut off match after end (there’s a bit after the match ends that’s dead data to us) (need more info from Ben to spot where to do so.)
• [ ] Get matches by index (there’s API for that) (may help us scale)
• [ ] Chart.js?
  • To get the scatter plot working, you need to set borderWidth: 0 in the dataset, I think? Otherwise it'll draw lines.

Later things:

• [ ] Official pro names (via appropriate API)
• [ ] Combatseries -> Allstate (?)
• [ ] Priority Celery queues for paying users
• [x] Replay parsing refactor: combatseries files reduced and combined into one, will allow combination directly with UpdateParseEnd
• [ ] Show filter criteria on match page / ??? to parsed?
• [ ] Screencast (for Ben to do)

Done things:

• [X] Re-keying / reducing all state size (see if there are existing JSON rekeying libraries?)
  • This doesn't gain us enough to be worthwhile.
• [X] EC2 replay URL service
  • Can run the replay URL service in docker, can push that to ECR/ECS. Need to finish testing side-by-side.
• [X] Autocompletes on match filter form (endpoints already exist for this?)

[bwarren2]

Other things I can think of:

Use js + rest API to get the status of a given match request on the match detail pages. This circumvents caching and allows us to add custom state messages based on the status of the match request.

Some way to periodically cycle the url backends? This might not even be taking down the process, just having the process log out of steam and log back in.

Disentangle a match parse from an API call. Right now, if someone tries to import a match, a chain of ValveApiCall + other things is set up; this makes the entire process block on the api_call queue, which is rate limited and can be slow. (Maybe factoring out the logic of Valve Api Call into a parent class that gets Subcriber and NonSubscriber implementations, potentially with different key sets and queues? This would matter when we have paying users, ie not now.) The tricky logic here is that IIRC the match parsing does rely on the logic from the VAC to know which heroes to look for etc; maybe a separate queue for VACs that are blocking for parsing?