Hey Martin
On Wed, Sep 28, 2022 at 4:18 PM Martin Urbanec martin.urbanec@wikimedia.cz wrote:
I also have one other question: Will scap sync-file stay with us? There are also complex deployments handled outside of backport windows that I think might benefit from it, as it offers fine-grained control on what is happening.
scap sync-file still exists.
But as we move towards Kubernetes we'll need to start thinking on the patch-level rather than the file-per-patch level. That is, what's merged is what will be deployed.
Apart from the dependencies between files within a patch, do you have specific complex scenarios in mind? (Feel free to reply here or in a task or on IRC—whatever's easiest for you).
Yes, you can do a +2 ahead of time and still use scap backport to do the remainder of the process, just be aware that if you do this and a patch ends up getting merged before the patch you are in the process of backporting with scap backport gets merged, then scap backport would end up syncing both patches at the same time (you will be given a warning and asked if you want to proceed).
Good to know, thanks! I tend to +2 all backports at once, and decide on the deployment order during the window. This means patches getting merged in the "wrong" order will likely be an issue. Are backports in different repositories still an issue in that regard? For example, if I have three patches for three extensions, +2 them at once, and then use scap backport, will it sync everything, or will it only fetch submodules
If you merge two changes, and then run `scap backport <one-change>`?
In that case, `scap backport <one-change>` will warn you that other changes have merged while you were trying to sync out `<one-change>`, and ask if you want to proceed with the sync—both changes will go out at the same time.
Now that we no longer have PHP revalidate files in the opcache and instead perform a restart: these changes should be atomic (mostly).
scap backport will always sync everything—not just submodules (this is fast now, I swear :)).
Good question...Do you mean the confirmation after syncing to debug servers? I assume you'd need to run the scripts right after pulling the changes to the deploy server. If so we would need to add an additional step.
I meant the "Would you like to see the diff" step, but there are many scripts, and the "when to run the script" differs. Some have to be executed as the very first step (createExtensionTables.php, for example). Other scripts need to be executed as the very last step (purgeList.php). There are also scripts that are not much sensitive to when they run, they just have to run at some point.
For createExtensionTables.php, you could +2 fetch down and run that: same as now—scap backport can do the sync when you're ready to proceed. We'll need to think about how new extensions are deployed in our Kubernetes future.
For things where you would like to run a check on migration changes, could you use the pause on mwdebug servers to do so? For these, would it be helpful to have a step to go to mwmaint servers, too?
For others, you can run them at the end—same as now.
Would that cover all the scenarios we have currently?
Thanks, as always, for being an amazing human and deployer
<3
– Tyler
Martin
út 27. 9. 2022 v 23:51 odesÃlatel Jeena Huneidi jhuneidi@wikimedia.org napsal:
Hi Martin, thanks for the feedback!
I often +2 all backports at the start of the window (or even prior the window), because I want to save time. It enables me to deploy more patches than I would be able to deploy with scap backport +2'ing it each patch separately. I also parallelize some of the steps (such as, +2'ing a config patch during the sync step), which also saves a bit of time. How would I do this with scap backport? By +2'ing the patch manually, and then letting scap backport figure out its magic?
Yes, you can do a +2 ahead of time and still use scap backport to do the remainder of the process, just be aware that if you do this and a patch ends up getting merged before the patch you are in the process of backporting with scap backport gets merged, then scap backport would end up syncing both patches at the same time (you will be given a warning and asked if you want to proceed).
Also, I'd like to mention that many ordinary deployments currently require a script to run. How would that work in scap backport world? When it asks me for confirmation, I will wait, run the script, and then finish it, to ensure the script runs at the right time
Good question...Do you mean the confirmation after syncing to debug servers? I assume you'd need to run the scripts right after pulling the changes to the deploy server. If so we would need to add an additional step.
Jeena
On Tue, Sep 27, 2022 at 2:23 PM Martin Urbanec martin.urbanec@wikimedia.cz wrote:
Hi,
Thanks for all the work on this project! I have some thoughts.
I often +2 all backports at the start of the window (or even prior the window), because I want to save time. It enables me to deploy more patches than I would be able to deploy with scap backport +2'ing it each patch separately. I also parallelize some of the steps (such as, +2'ing a config patch during the sync step), which also saves a bit of time. How would I do this with scap backport? By +2'ing the patch manually, and then letting scap backport figure out its magic?
Also, I'd like to mention that many ordinary deployments currently require a script to run. How would that work in scap backport world? When it asks me for confirmation, I will wait, run the script, and then finish it, to ensure the script runs at the right time
Martin
út 27. 9. 2022 v 23:05 odesÃlatel Tyler Cipriani tcipriani@wikimedia.org napsal:
tl;dr: use `scap backport <gerrit url>` to deploy MediaWiki backports ____
There’s now a single step to deploy changes to Wikimedia’s production MediaWiki.
🤯 An 85% reduction in command remembering!
On the deployment host run:
scap backport <gerrit url>This works for any change to a live branch for mediawiki/core, extensions, skins, or operations/mediawiki-config. ____
More details (and a demo) inside Jeena Huneidi’s excellent write up.
<3
– Tyler Cipriani (on behalf of the RelEngers who really do make dreams come true) Engineering Manager, Release Engineering Wikimedia Foundation _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
-- Jeena Huneidi Software Engineer, Release Engineering Wikimedia Foundation _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/