[meta] Migrate `mach try` submission to Lando
Categories
(Conduit :: Lando, enhancement)
Tracking
(Not tracked)
People
(Reporter: sheehan, Assigned: sheehan)
References
(Depends on 13 open bugs)
Details
To alleviate long wait times experienced when pushing a commit to the try repository we will update mach try to submit try requests to Lando. Lando will use existing queuing and working infrastructure to push the commit to the repository.
Comment 1•2 years ago
|
||
Is this going to mean that try pushes will be subject to the commit number limit for lando landings? That limit currently makes the wpt sync bot unable to use lando for its operation.
(In reply to James Graham [:jgraham] from comment #1)
Is this going to mean that try pushes will be subject to the commit number limit for lando landings? That limit currently makes the wpt sync bot unable to use lando for its operation.
This work is only for human initiated try pushes, in part because of the SSO authentication requirements. The ssh based workflow is not being removed and should continue to be used by automation.
If/when we route all try submissions through Lando we'll have accommodations for bots; we're already mindful of the size of the wpt stacks 🙂. There's no plans to work on this aspect right now.
| Assignee | ||
Updated•2 years ago
|
Comment 3•2 years ago
|
||
The ssh based workflow is not being removed and should continue to be used by automation.
What's this expected to look like command line API-wise? i.e. will there be some additional flags to e.g. mach try fuzzy required to get the current behaviour, or will it somehow autodetect when it's possible to route the push via lando?
| Assignee | ||
Comment 4•2 years ago
|
||
(In reply to James Graham [:jgraham] from comment #3)
The ssh based workflow is not being removed and should continue to be used by automation.
What's this expected to look like command line API-wise? i.e. will there be some additional flags to e.g.
mach try fuzzyrequired to get the current behaviour, or will it somehow autodetect when it's possible to route the push via lando?
I'm expecting the Lando-based push to become the default, and a flag/environment variable which can force the push directly to hgmo.
Comment 5•2 years ago
|
||
To alleviate long wait times experienced when pushing a commit to the try repository
Not saying anything about whether the mach try move to Lando should or should not happen, but I'll just note that if anything, wait times will only increase with the change. What will change however, is that the wait time on the user command line will disappear. But the wait time between running mach try and having something actually visible on treeherder will be longer.
(In reply to Mike Hommey [:glandium] from comment #5)
Not saying anything about whether the mach try move to Lando should or should not happen, but I'll just note that if anything, wait times will only increase with the change. What will change however, is that the wait time on the user command line will disappear. But the wait time between running mach try and having something actually visible on treeherder will be longer.
The goal of this work has always been to address just the command-line wait time.
It will not impact the total end-to-end times by any meaningful amount in either direction.
The bulk of the time waiting at the command line is waiting for the global lock on the repository; this will now happen in Lando rather than in each engineer's terminal.
Comment 7•2 years ago
|
||
One feature that would be nice here would also be the ability to cancel, while in lando, a pending try push.
This can happen sometimes when you do a try push from the wrong revision, and the current story requires waiting for the decision task to have run on your push, then cancelling it. It would be lovely if you could catch it earlier in the lando phase and nip it there.
(In reply to Matthew Gaudet (he/him) [:mgaudet] from comment #7)
One feature that would be nice here would also be the ability to cancel, while in lando, a pending try push.
That's bug 1652787.
Updated•2 years ago
|
Updated•2 years ago
|
Oops, wrong bug.
Updated•2 years ago
|
Comment 10•1 year ago
|
||
Moving this out from blocking hg-to-git production deployment, after discussion with Connor and the team.
Comment 11•1 year ago
|
||
Updatebot currently uses --push-to-vcs and will take some time to migrate to --push-to-lando; before push-to-vcs goes away forever we would appreciate a head's up with as much lead time as possible to figure out who can implement the fix (and what actually needs to be changed.)
Updated•7 months ago
|
Description
•