We recently ended up shutting down our production Hasura deployment, because we were only using it for data ingestion, and we ended up in a situation where Hasura was constantly using all available memory, and then crashing.

I don't have any logs to offer, unfortunately. But I was asked on Discord to create an issue, so here's the information I can provide about our setup:

• Deployed on Heroku
• Initially ran on a $7/mo box on v1alpha1 build 28 (We didn't watch it closely, and at some point realized that it was crashing regularly from running out of memory)
• Scaled the box to larger standard dynos, even up to 3 boxes with 2.5Gb of ram each. Even in that situation, Hasura used all and more than the available ram, and was killed by Heroku.
• Heroku's router measurements said we were getting about 10 reqs / second from our client devices.
• The client devices were doing mutations only. No queries. They were writing in telemetry events.
• Sometimes queries could be multiple megabytes in size (a lot of events were in the write backlog)
• We used an auth hook to our application server that responded usually within a few hundred milliseconds.
• Sometimes the auth hook would throw a 500 error.
• Right before crashing the Hasura logs showed mutation requests timing out (taking longer than 30 seconds)
• Rebooting the box helped for a few seconds, but the slowness almost immediately returned.
• We truncated our events table to make sure the database wasn't the bottleneck, but the memory issues on Hasura continued.

Sorry I can't offer more. Thank you!