After #2247 active TCP connections per host are limited to avoid things blowing up. While testing #2254 with a network limiter I spotted a case where one build was ongoing and doing a pull, other one couldn't really start because the image config resolve in Dockerfile was stuck in a queue blocking the whole build until the layers from the other build completed.
I wonder if it would make sense to have certain high-priority requests like HEAD and GET for manifest and config blobs that could get a single extra connection. These requests are small and don't take much time but they may unblock big parts of a build.
Not sure how this technically can be achieved with the current use of x/sync/semaphore package.
@sipsma @aaronlehmann
After #2247 active TCP connections per host are limited to avoid things blowing up. While testing #2254 with a network limiter I spotted a case where one build was ongoing and doing a pull, other one couldn't really start because the image config resolve in Dockerfile was stuck in a queue blocking the whole build until the layers from the other build completed.
I wonder if it would make sense to have certain high-priority requests like HEAD and GET for manifest and config blobs that could get a single extra connection. These requests are small and don't take much time but they may unblock big parts of a build.
Not sure how this technically can be achieved with the current use of
x/sync/semaphorepackage.@sipsma @aaronlehmann