A variety of errors are happening, at a 5-15% rate, meaning that many callers get an error at some point during their call.
Some basic details:
Lots are timeouts (no response within 15 seconds, so Twilio kills it) — these could be spin-up time if any of the instances are on the free, sleeping tier. Someone should confirm none are allowed to sleep.
A number are 502s (upstream bad response) — not sure what those are about.
Some (probably a plurality based on my limited sampling) of the errors occur at the initial call, but many occur a number of interactions into the call (say, at question 3) so it's unlikely to be something just related to the core URL.
I've added basic 7-day log persistence (free tier) to most of the instances ( #94 will add it to all of them) so investigating the Heroku app-level logs and cross-referencing the Twilio errors.
Another thing to doubly check — I know some instances have been renamed. We should make sure all Twilio phone numbers have been updated to point at the new name (this doesn't happen automatically.)
Moving forward, a good process would be to have a Twilio error alert set up (easy via the admin console) for the cityvoice-builder subaccount, so that escalated error rates reach someone proactively.
I've reached out to a Rails person I trust who has a lot of experience with Heroku to get a copy of his Puma config settings — we should move to that on these, see if it reduces/eliminates the errors, and if so make that change permanent in master.
A variety of errors are happening, at a 5-15% rate, meaning that many callers get an error at some point during their call.
Some basic details:
I've added basic 7-day log persistence (free tier) to most of the instances ( #94 will add it to all of them) so investigating the Heroku app-level logs and cross-referencing the Twilio errors.
Another thing to doubly check — I know some instances have been renamed. We should make sure all Twilio phone numbers have been updated to point at the new name (this doesn't happen automatically.)
Moving forward, a good process would be to have a Twilio error alert set up (easy via the admin console) for the cityvoice-builder subaccount, so that escalated error rates reach someone proactively.
I've reached out to a Rails person I trust who has a lot of experience with Heroku to get a copy of his Puma config settings — we should move to that on these, see if it reduces/eliminates the errors, and if so make that change permanent in master.