Lots of database writes once an act_runner is registered and running
Opened this issue · 6 comments
Description
As soon as there is an act_runner (v0.2.11) running (even if it is idle), there appears to be a constant flow of database updates.
As far as I can tell, this is due to the last_online field in the act_runner table being updated all the time.
By default this happens every two seconds for each act_runner. I would much prefer if there were no database writes while the runners are idle. I couldn't find a way to disable this behavior. It would be nice to have an option to disable this or possibly keep the last_online value in memory only instead of writing it to the database.
Thanks for the great work on Gitea!
Gitea Version
1.23.5
Can you reproduce the bug on the Gitea demo site?
No
Log Gist
No response
Screenshots
No response
Git Version
2.48.1
Operating System
Alpine Linux (edge)
How are you running Gitea?
I'm using the Gitea package provided by Alpine Linux.
Database
SQLite
Curious about this as well. Realised the hdd was just constantly going and have tracked it down to this file. Would like a quieter system when idle.
This isn’t a bug — it’s the intended behavior. That said, I agree it’s not a great design. The key question is: what’s the actual impact of this behavior?
@lunny Thanks for the reply. As their is a call to the Gitea server each time, it makes for very loud logs. This makes it hard to investigate issues looking through those logs. My assumption would there is good log rotation so they aren't slowly eating up storage space? Constant log writing and database editing is going to be a bit wearing on the hardware I would imagine (though probably fairly minor).
I'm fully self-hosted (that is, a server in my living room) so while excess writes and storage isn't a crucial issue, it's a bigger deal for my situation than if I was using a cloud service provider who will swap the hardware for me in a timely (and experienced!) manner.
I'm not sure if a potential solution would be allowing the timing to be customised (feels like kicking the can down the road) or a different design.
I hope that helps! If my assumptions/worries are baseless, feel free to correct them!
This isn’t a bug — it’s the intended behavior. That said, I agree it’s not a great design.
What's the purpose of those database updates? Is it just for the user to be able to see when the act_runner last talked to Gitea or is that information used for something else as well? As far as I can tell, Gitea doesn't really need this information since the act_runners actively ask for new tasks anyway, but maybe I am missing something here.
Constant log writing and database editing is going to be a bit wearing on the hardware I would imagine
It is and on SSDs it is actually quite a bit more than what one might expect due to write amplification of SSDs and due to the fact that each database write - no matter how small it is - is followed by an fsync call. Of course using fsync makes perfect sense for a database to ensure its consistency, but that also means that the OS can't batch all those small write into fewer larger ones. Neither can the SSD do it.
I also noticed that the SSD in my server runs several degrees hotter with an act_runner active, probably because the SSD can never enter its power saving states due to the constant stream of writes.
Without act_runners active, Gitea appears to write very little to disk, if nothing is going on, which is nice.
Yes, just store the last active time. Maybe the state can be stored in memory/redis.
I just came across this same issue. My mini home server that hosts my git forge is on my shelf right next to my desk in my office, and when its 4 hard disks are active its noticeably loud. Yesterday I enabled a runner, and since then my server has been spinning up its disks every 2 seconds to write to the sqlite db. Its very distracting, and I feel like it can't be good for my disks.
Is there a way to disable this "last online" update? or store it in memory and commit to disk on shutdown, or even once every 10 minutes?
For now, I've just shut down my runner. I'll turn it back on only when I need to use it.