feat(server): Use a single tokio runtime, adjust processor task count #3516
Conversation
Dav1dde
force-pushed
the
dav1d/remove-runtimes
branch
2 times, most recently
from
bedbba1 to
bf1ed87
Compare
Dav1dde
force-pushed
the
dav1d/remove-runtimes
branch
from
bf1ed87 to
8c3d622
Compare
|
We can, but I don't see this as risky and would have moved on from canary and S4S after a workday. |
| @@ -586,7 +586,7 @@ struct Limits { | |||
| max_replay_message_size: ByteSize, | |||
| /// The maximum number of threads to spawn for CPU and web work, each. | |||
| /// | |||
| /// The total number of threads spawned will roughly be `2 * max_thread_count + 1`. Defaults to | |||
| /// The total number of threads spawned will roughly be `2 * max_thread_count`. Defaults to | |||
There was a problem hiding this comment.
Can we add to this comment that Relay will use at most max_thread_count CPU?
There was a problem hiding this comment.
With the current setup this isn't gurantueed right? If the runtime needs more processing power than expected it can go over.
| 0 | 1 => 1, | ||
| 2 | 3 => 2, | ||
| 4 => 3, | ||
| conc => conc - 2, |
There was a problem hiding this comment.
Have you run load tests with this? It would be good to know the actual needs here, if this is easy to do.
There was a problem hiding this comment.
No, this was simply a best effort guess for small amounts of cores going off our discussed -2 starting point.
Having a small amount of cores would reserve a much bigger % to the runtime than if we have a large amount of cores. E.g. for 2 cores allocating 1 thread to the processor would allocate 50% to the runtime, while having 12 cores only allocates 17% to the runtime.
As described in #3513.
Will be followed by an ops PR to adjust configuration settings.