Launchpad Mojo Specs

~cjwatson/launchpad-mojo-specs:lp-codeimport-ps5

  1. Git
  2. lp:~cjwatson/launchpad-mojo-specs
  3. lp-codeimport-ps5
Last commit made on 2021-06-09
Get this branch:
git clone -b lp-codeimport-ps5 https://git.launchpad.net/~cjwatson/launchpad-mojo-specs
Only Colin Watson can upload to this branch. If you are Colin Watson please log in for upload directions.

Branch merges

Branch information

Name:
lp-codeimport-ps5
Repository:
lp:~cjwatson/launchpad-mojo-specs

Recent commits

52a3dfd... by Colin Watson

WIP

b479856... by Paul Collins

decommission oci-webhooks.admin.canonical.com

09e722c... by Ioana Lasc

git: upgrades turnip-celery charm

0cb53e2... by Colin Watson

signing: Allow bilimbi to use internal API

This is needed for signing copy archives.

da0a4c5... by Ioana Lasc

git: Upgrade turnip-celery charm

0bcdeb3... by Ioana Lasc

git: Upgrade turnip-celery charm

badab31... by Ioana Lasc

git: Add dedicated repack celery worker

571f506... by Colin Watson

Merge branch 'git-hard-mount'

7e2cc6b... by Colin Watson

git: Use "hard" NFS client recovery

From nfs(5):

  soft / hard Determines the recovery behavior of the NFS client after
  an NFS request times out. If neither option is
  specified (or if the hard option is specified), NFS
  requests are retried indefinitely. If the soft option
  is specified, then the NFS client fails an NFS request
  after retrans retransmissions have been sent, causing
  the NFS client to return an error to the calling
  application.

  NB: A so-called "soft" timeout can cause silent data
  corruption in certain cases. As such, use the soft
  option only when client responsiveness is more important
  than data integrity. Using NFS over TCP or increasing
  the value of the retrans option may mitigate some of the
  risks of using the soft option.

I think I initially went for "soft" because retrying forever rather than
failing seemed undesirable, and because the code writing to this NFS
mount is very limited and known to have reasonable error handling.
However, it seems that under load this can also cause the client to
consider the NFS mount dead after relatively short periods of
unresponsiveness when it might otherwise be able to recover. Let's see
if using the default of "hard" works more gracefully in practice.

LP: #1928140

7e8b43a... by Ioana Lasc

git: Add charms back to collect-charm-upgrades