Speeding up build/run requests with a pool of idle docker containers #463
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ffissore
force-pushed
the
docker_pool
branch
from
deeddb0 to
644c0f9
Compare
|
Also, this fix is for |
|
Thank you! As a fair warning, it will probably take me a little while to dig in here. If I haven't usefully commented in a few weeks, feel free to ping me again! |
|
As requested, a kind ping to check this PR. I've just rebased it with the latest master |
started and idle docker containers, rather than starting a new one for each request. By sleeping for a long time in entrypoint.sh, containers stay idle. Cargo is called through cargo.sh, which cares of exporting env vars and killing the container once it's done. Thus, there's no need to drop a container that has been used. However idle containers need to me manually terminated: DockerContainers Drop impl takes care of that.
the playground Also containers are now stopped in parallel
|
Hello there, another kind ping to check this PR. I've just rebased it with the latest master |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is a possible solution to #457
I've sped up the compilation/execution of snippets by using previously started and idle docker containers, rather than starting a new one for each request.
On my dev box, this new version runs a "hello world" in ~600 ms, while the master version takes ~1.3 secs. Since
rustctakes ~400 ms, the overhead of docker is reduced to ~200 ms per request.How it works
A pool of idle docker containers is started at playground start up. The default size is 10 containers per channel.
By sleeping for a long time in
entrypoint.sh, containers stay idle.When a request is handled, a container is removed from the pool. Cargo is called through
cargo.sh, which cares of exporting env vars and killing the container once it's done. Thus, there's no need to drop a container that has been used.However idle containers in the pool need to me manually terminated: DockerContainers
Dropimpl takes care of that, and kills them in parallel.To ensure that
dropis called when server shuts down, I've used cratectrlcto install a shutdown hookKnown issues
Benchmarking the
/executeapi withsiege, the new pool version is slightly slower than the master one: the former replies to ~4.5 req/sec, the latter to ~5.5 req/sAlso, because there's no way to gracefully shutdown iron (see hyperium/hyper#338), if the server is terminated while receiving lots of requests, there's a chance that some docker containers will survive the shutdown: they'll have to be manually terminated.
On a side note, I'm new to rust, so any feedback is welcome