search

goodrobots / maverick-api

API backend for maverick-web
MIT License
8 stars 5 forks source link

Low-level websocket error when connecting to -api #228

Open fnoop opened 4 years ago

fnoop commented 4 years ago

Looks like when initial subscription is made the websocket gets cut off or interrupted. Is it from the server side? Can't see anything obvious in the logs.

Screenshot 2020-04-08 at 08 18 42 Screenshot 2020-04-08 at 08 19 19
SamuelDudley commented 4 years ago

The top one is new to me. Do you think it has something to do with the client attempting to connect to /subscriptions before the schema has loaded? Will investigate further on -api side.

I have certainly seen the bottom one before. I think it's related to the browser trying to reconnect to a websocket -api has closed. Possibly -api is closing them incorrectly. I'll investigate.

fnoop commented 4 years ago

Sorry should have been clearer - they're both the same thing but seen in different browsers. The top one is firefox, the bottom one is chrome. This happens when the page is first loaded, and the clients are first loaded. It's happening right down at the apollo client level. We've changed all the client creation to async, but it doesn't try to create the initial query/subscription until the schema has been fully downloaded and parsed:

          // If schema has been fetched, proceed to create Status query/subscription
          if (newValue[api].schemaready === true) {
            this.createQuery('Status', statusQuery, api, null, null, this.processStatusQuery)
            this.createSubscription('Status', statusSubscription, api, null, null, this.processStatusSubscription)
          }

It's entirely possible it's something to do with timing at the client end..

SamuelDudley commented 4 years ago

Okay. If you close the web browser, restart -api and then load the page do you see the error?

fnoop commented 4 years ago

No, although actually if I hit reload lots of times it happens sometimes and not others. Odd. I guess that probably points to a brower-side timing issue with the async calls?

SamuelDudley commented 4 years ago

I actually think it's to do with -api and websocket connections not being disposed of correctly. There is an error that occurs in the logs I see every now and then generated from here https://github.com/goodrobots/maverick-api/blob/master/maverick_api/modules/base/tornadoql/subscription_handler.py#L211