-
Notifications
You must be signed in to change notification settings - Fork 1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
lcow: Slower pull times with newer kernels #3120
Comments
Trying to repro the issue with:
The daemon is started on a fresh
A quick test with
Let's look at it more systematically and use both
By repeating the For
For
For
So, there is indeed a significant difference between the two kernels. Let's bisect... |
We use the
So the regression seems to have happened between |
As a quick check I looked at boot times of 4.14.35 and 4.14.36 kernels using a quick hack |
Between 4.14.35 and 4.14.36 some patches were introduced which slow down image load/pull times considerably. For now force downgrade the kernel See: linuxkit/linuxkit#3120 Signed-off-by: Rolf Neugebauer <rolf.neugebauer@docker.com>
May be coincidental, but 4.14.36 contained some ext4 correctness fixes as well as a change to the locking in writeback. I only mention this because I had to do some looking through that same changelog in order to figure out what was going on with random number generation: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/log/?h=v4.14.36 |
yeah, I saw that too. Haven't gotten around bisecting this one yet. |
Closing now it's bumped to 4.19.x. Doesn't seem to repro any more. |
Oh I don't seem to have permission to close my own issue in this repro.... but @rn or @justincormack probably can. Edit: Aha, Rolf opened it on my behalf. Should have looked at the top :) |
This was reported by @jhowardmsft on slack. He noticed that with newer kernel, specifically 4.14.53, the first pull on fresh data root is a lot slower than on older kernels, specifically he tested 4.14.23. Apparently the
mkfs.ext4
seems to take a lot longer. One suspicion is the inclusion of0020-scsi-storvsc-Allow-only-one-remove-lun-work-item-to-.patch
and0021-scsi-storvsc-Avoid-excessive-host-scan-on-controller.patch
, which were introduced with 4.14.30 as part of some other debugging effort around LCOW and disk hotplug.This is issue is to track the debug
The text was updated successfully, but these errors were encountered: