didn’t see this type of topic on the forum, would like to ask what’s the plan of supporting logging and metrics (and alarms) in the ICP world. All of them are important for monitoring healthy of services.
6 Likes
What do you think the plan should be?
I want:
-
dfx logsto stream logs from all the canisters interleaved, but prefixed/color-coded by canister name. I can pipe togrep. Could be cool to support some kind of streaming json export I can pipe tojq- inspiration: heroku used to recomend foreman(1) - manage Procfile-based applications + Procfile.
lerna rundoes this.
- inspiration: heroku used to recomend foreman(1) - manage Procfile-based applications + Procfile.
On the other hand, this is tricky. If logging takes more cpu/mem, should everyone have to pay for it? Can logging be in userspace via some kind of standard API on the canister?
Hi bengo,
Apologize for the delay, somehow I missed your response.
dfx logsto stream logs from all the canisters interleaved, but prefixed/color-coded by canister name. I can pipe togrep. Could be cool to support some kind of streaming json export I can pipe tojq
Yes, I like the idea. It is similar to AWS cloudwatch that multiple compute instances output their logs to one single log stream. Also, it has the query functionality over the logs (analogy to the using `jq`` over the exported logs)
On the other hand, this is tricky. If logging takes more cpu/mem, should everyone have to pay for it? Can logging be in userspace via some kind of standard API on the canister?
can you elaborate on Can logging be in userspace via some kind of standard API on the canister? ?
If logging takes more cpu/mem, should everyone have to pay for it?
If the logging incurred by an API invocation, it makes sense the fee of invoking the API considers including the logging cost. If logging is produced by systems, it might make sense charge the cost on the owner of the canister.
Thanks!