1324052 - add macosx64 nightly builds to the taskgraph

Status: RESOLVED FIXED

(Firefox Build System :: Task Configuration, task)

Comment 1

[Kim Moir [:kmoir] ET]

Attachment: bug1324052.patch

https://treeherder.mozilla.org/#/jobs?repo=try&revision=e2114e3abe5d3d3f2803edfbfddf2576c938a872

Comment 2

[Kim Moir [:kmoir] ET]

Attachment: bug1267425-2.patch

Comment 3

[Kim Moir [:kmoir] ET]

Attachment: bug1267425-3.patch

https://treeherder.mozilla.org/#/jobs?repo=try&revision=c55c54c52bdd23dc24f242fbc6bb27b91f270af8

Comment 4

[Kim Moir [:kmoir] ET]

Comment on attachment 8819399 [details] [diff] [review]
bug1267425-3.patch

I frustrated with the mach repackage bug so I decided to work on another bug instead. I realize that all the dependencies for this graph are not implemented but thought it would be useful to implement this in the taskgraph on date, it won't be invoked until the hook is added anyways.

Comment 5

[Justin Wood (:Callek)]

Comment on attachment 8819399 [details] [diff] [review]
bug1267425-3.patch

Review of attachment 8819399 [details] [diff] [review]:
-----------------------------------------------------------------

I'm working on the tc-l10n bits now locally, not sure if they will be this easy...

That said, I'm aware of some gotchya's in how we did l10n so far, so probably easiest to leave l10n out for now (but I don't actually mind if it lands)

Secondarily my other comments are available for debate/discussion...

Overall this is very good (I didn't run locally to check its task output, though)

::: taskcluster/ci/build/macosx.yml
@@ +53,5 @@
> +        product: firefox
> +        job-name: macosx64-nightly-opt
> +    treeherder:
> +        platform: osx-10-7/opt
> +        symbol: tc(B)

tc(N)

::: taskcluster/ci/nightly-l10n/kind.yml
@@ +132,5 @@
> +        treeherder:
> +            platform: macosx64/opt
> +        worker:
> +            env:
> +                EN_US_PACKAGE_NAME: target.tar.bz2

target.dmg

::: taskcluster/taskgraph/target_tasks.py
@@ +158,5 @@
>              return task.attributes.get('nightly', False)
>      return [l for l, t in full_task_graph.tasks.iteritems() if filter(t)]
> +
> +
> +@_target_task('nightly_macosx')

I'm of the mindset that we should have all desktop as one target_tasks method, and not per-platform on desktop.  This is of course debatable.

Comment 6

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-4.patch

Thanks for the feedback

I was going to leave the macosx part as a separate call so the current hook doesn't initiate it until we have the dependent issues fixed.

Comment 7

[Justin Wood (:Callek)]

Comment on attachment 8819421 [details] [diff] [review]
bug1324052-4.patch

Review of attachment 8819421 [details] [diff] [review]:
-----------------------------------------------------------------

::: taskcluster/ci/nightly-l10n/kind.yml
@@ +144,5 @@
> +                - taskcluster_nightly.py
> +            options:
> +                - environment-config=single_locale/production.py
> +                - branch-config=single_locale/{project}.py
> +                - platform-config=single_locale/linux64.py

last minute catch s/linux64/macosx64/

Comment 8

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-5.patch

Comment 9

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-6.patch

Comment 10

[Kim Moir [:kmoir] ET]

https://hg.mozilla.org/projects/date/rev/5a08d80e3b3f411f261fd2819d913ccbfdc255d9

Comment 11

[Kim Moir [:kmoir] ET]

Attachment: bug1324052nightly.patch

patch to consolidate nightly builds once mac builds are ready

hook definition will have to be updated to from nightly_linux to nightly_all

Comment 12

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-7.patch

unbitrotten patch.  I still have to refactor it to account for the signing and packaging pipeline in bug 1318505

Comment 13

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-8.patch

Comment 14

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-9.patch

So this is a patch I have in progress for nightly mac builds which fixes up the artifact format etc.  It doesn't have the mach repackage task added

https://treeherder.mozilla.org/#/jobs?repo=try&revision=1de627b8fe2e15e72b512c7b82cf7e5a35b39d20

I realize the next step after running the signing job is is take the tarball it creates and pass it to mach repackage task to create the dmg with the signed artifacts.

I'm unsure of the parameters for the mach repackage job.  Is it just a build task with different parameters than the first unsigned build job?

Comment 15

[Kim Moir [:kmoir] ET]

notes: I haven't fixed taskcluster/taskgraph/transforms/beetmover.py yet, it's a todo

Comment 16

[Justin Wood (:Callek)]

So, the (In reply to Kim Moir [:kmoir] from comment #14)
> Created attachment 8845653 [details] [diff] [review]
> bug1324052-9.patch

This looks like a good direction

> I realize the next step after running the signing job is is take the tarball
> it creates and pass it to mach repackage task to create the dmg with the
> signed artifacts.
> 
> I'm unsure of the parameters for the mach repackage job.  Is it just a build
> task with different parameters than the first unsigned build job?

It will be a mozharness job, probably a different script.  The main reason for even using mozharness is to gather the tooltool bits to do the dmg creation and running configure.

I'm not certain how this will all look yet, but it will need the url to the signed tarball of course, and need chain of trust enabled.

this is the job that should also have the signing related routes attached, because it will be the deliverable we get out of signing.

Hope that helps.

Comment 17

[Kim Moir [:kmoir] ET]

Attachment: wip patches

sort of fell down a refactoring rabbit hole this afternoon. These patches are a wip, don't work yet.

Comment 18

[Kim Moir [:kmoir] ET]

I went for a walk in the snowstorm tonight and was thinking about my code.  I think I took the wrong approach with the part where I'm trying to transform the build-signing task to a repackage task.

I think I need to just add a new task that, as aki suggested in irc

task.payload.upstreamArtifacts will point at the mac build-signing task
task.payload.features.chainOfTrust = True
and transform some of the upstream attributes to be a new repackage task. 
fix routes

I'm not sure of the scopes, the worker type, or the mh script to be called.

Comment 19

[Kim Moir [:kmoir] ET]

Attachment: old wip patches

Comment 20

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-3.patch

patches from this morning, refactored on top of bug 1333255, going to define a new task implementation for repackage now

the existing task implementations of balrog, beetmover, and scriptworker-signing have upstream artifacts defined but this task doesn't really fall in these categories

Comment 21

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-4.patch

running a try run now, not quite ready yet but making progress

Comment 22

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-5.patch

Try run is here
https://treeherder.mozilla.org/#/jobs?repo=try&revision=8f4ea23a3966383ca415f72eb4aa0494c4835d8a

https://public-artifacts.taskcluster.net/BKXYAlt2SLaW-GuRkbvfVA/0/public/full-task-graph.json

has task repackage-macosx64-nightly

things I still need to add but not sure of values
scopes
mh script to call

Also, not sure how to enable cot as feature in payload

will ask in #tcmigration

Comment 23

[Aki Sasaki (not active)]

- I think we should enable run-on-projects once we're ready
- I don't think scopes are important for this
- I don't know if the mh script is written yet?
- to enable cot generation, set task.payload.features.chainOfTrust to True

Comment 24

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-6.patch

I tried several different ways last night and this morning
to enable cot generation, set task.payload.features.chainOfTrust to True

and it didn't work (wasn't included in graph). So if you have suggestions, I would appreciate it.  Also, I'm not sure what the worker-type should be, do we need to define a new one or just use the build one which is gecko-1-b-linux?

Comment 25

[Aki Sasaki (not active)]

Is this what you're looking for?
https://hg.mozilla.org/integration/mozilla-inbound/file/tip/taskcluster/taskgraph/transforms/task.py#l461
https://hg.mozilla.org/integration/mozilla-inbound/file/tip/taskcluster/taskgraph/transforms/build.py#l25

You probably want to follow the build transform's chain of trust settings, including the docker image builder in task.extra.chainOfTrust.inputs as well.

I'm guessing the build workerType is best.

Comment 26

[Kim Moir [:kmoir] ET]

Yes, I tried that approach above but it didn't work but I will play with it some more.  thanks!

Comment 27

[Aki Sasaki (not active)]

If you're talking about

+        task = {
+            'label': label,
+            'description': "{} Signing".format(
+                dep_job.task["metadata"]["description"]),
+            'worker-type': "scriptworker-prov-v1/signing-linux-v1",
+            'worker': {'implementation': 'repackage',
+                       'upstream-artifacts': job['upstream-artifacts']},
+       #     'scopes': [signing_cert_scope] + signing_format_scopes,
+            'scopes': repackaging_format_scopes,
+            'dependencies': {job['depname']: dep_job.label},
+            'attributes': attributes,
+            'run-on-projects': dep_job.attributes.get('run_on_projects'),
+            'treeherder': treeherder,
+            'routes': job.get('routes', [])
+        #    'features': {'chainOfTrust': 'true'}

you commented out adding task.features, rather than setting task.worker.chain-of-trust

Comment 28

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-7.patch

Aki, I still can't seem to get the chain of trust bits in the graph and it would be great to have a second set of eyes as to what I'm doing wrong.

Comment 29

[Aki Sasaki (not active)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=875e18131190

Comment 30

[Aki Sasaki (not active)]

Ok.  A few things:

I pushed my changes to https://hg.mozilla.org/try/pushloghtml?changeset=4cbd3683dc5ee31e45f04d5d7e025d23504b9b31 .
I was able to get task.payload.features.chainOfTrust set to True using the nightly_macosx target_task_method locally.

- we don't need both transforms.repackage and transforms.repackaging.  I combined them and got the same graph

- the changes we make to the transforms.repackage task or job are basically changes that we pick up in transforms.task.build_repackage_payload .  So if you add something like task['chain-of-trust'], you have to change the task definition in that function, and remove it from the task dict.  If all tasks of a certain type have a certain feature, you can hardcode it there, or in the transforms, or taskcluster.ci yml files.

- upstreamArtifacts and scriptworker scopes are only relevant to scriptworker.  We're going to be running mac repackaging on docker-worker, so this is going to look a lot more like the build kind than the signing kind.  docker-worker requires a certain payload schema [1].  Most likely you want to pass through the @payload_builder('docker-worker') in transforms.task.

Does that help?

[1] https://github.com/taskcluster/docker-worker/blob/master/schemas/payload.json

Comment 31

[Kim Moir [:kmoir] ET]

yes, the taskgraph looks good thank you very much. Sorry I got off on the wrong path with two transforms, I tried the single transform earlier and I couldn't get it to work.  I should have asked more questions earlier.

Comment 32

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-8.patch

patches cleaned up to removed unused wip, fix pep8 errors

Comment 33

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-9.patch

More cleanup. 

https://treeherder.mozilla.org/#/jobs?repo=try&revision=f3c9d8f607d57750cc2ecfc07f5b9ee7f478e34c

Now that the graph is created correctly, I could land these on date (note the the patches are against m-c, may have to rebase a bit against date).

I talked to Callek, we are still sorting out how the mh script will be called, I will add an env to do that next

Comment 34

[Aki Sasaki (not active)]

Comment on attachment 8848507 [details] [diff] [review]
bug1324052tc-9.patch

You're still setting upstreamArtifacts (scriptworker only); I think these will break docker-worker.  It also looks like you're still missing a bunch of the docker-worker task definition.

Here's a docker worker linux build from date:
https://queue.taskcluster.net/v1/task/azRfkWWjT7e7FAjBu3OaRg

The repackage task definition should look similar to that, with a different command and env.

Comment 35

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-10.patch

updated patches. still some work to be done.

Comment 36

[Kim Moir [:kmoir] ET]

Attachment: sortedmac.json

So the previous patches used implementation of docker-worker which isn't correct.  I tried to change it to use the repackage implementation but it's not working and I think I need to look at it in the morning with fresh eyes. Going to ask aki if we can look at this together if I can't find solution it in the morning.

The attached file is the sorted json that is created with the previous patches.  The scopes should be removed.  I don't know if relengAPIProxy or taskclusterProxy are required. The mh command is not really defined yet because I'm not sure of the parameters and it needs a env to point to the signed artifacts according to Callek.

I'm not sure if the workertype is correct either.

Comment 37

[Kim Moir [:kmoir] ET]

Attachment: latest.patch

latest patch doesn't work

I want to have

 'worker': {'implementation': 'repackage',

in taskcluster/taskgraph/transforms/repackage.py

and it doesn't work

Comment 38

[Kim Moir [:kmoir] ET]

So here is the conclusion from talking to to aki

I should switch to transforming the build kind and override with the repackage kind.  It includes both mh and docker-worker.  So I should be able to transform it to include the correct command and env. The env should include a link to the task that will have the signing artifacts.  

kmoir-mtg: 'dependencies': {job['depname']: dep_job.label},
1:55 PM "https://queue.tc.net/.../{depname:label}/..."

When the optimization of the graph completes, the repackage task will have a url to the task of the signed artifacts.

Comment 39

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-11.patch

This patch is making progress. I'm attempting to replicate the bits from Callek's task here 
https://tools.taskcluster.net/task-inspector/#Av6XRz0qT76mdO9C3HnsZQ/

My try run
https://treeherder.mozilla.org/#/jobs?repo=try&author=[e-mail]&selectedJob=85479249

task graph generated from try run with repackage-macosx64-nightly/opt
https://public-artifacts.taskcluster.net/TRfl3l3PQt-tKqV_FM5HsA/0/public/full-task-graph.json

The things I am missing are 

relengAPIProxy - callek suggested looking at 
https://dxr.mozilla.org/mozilla-central/source/taskcluster/taskgraph/transforms/job/mozharness.py#130
https://dxr.mozilla.org/mozilla-central/source/taskcluster/taskgraph/transforms/job/mozharness.py#50

but I couldn't get it to work

The SIGNED_INPUT does not have the correct value, I think I know how to fix this
"SIGNED_INPUT": "https://queue.taskcluster.net/{depname:label}/target.tar.gz",

this is missing
"artifacts": {
    "public/build": {
      "path": "/home/worker/workspace/build/upload/",
      "expires": "2017-04-17T19:18:20.235910Z",
      "type": "directory"
    }
  },

Anyways, I will look at it again in the morning

Comment 40

[Kim Moir [:kmoir] ET]

Also, the dependencies are wrong

"dependencies": {
      "build": "build-macosx64-nightly/opt"
    },

They are fine if you look in the repackage task
 u'dependencies': {u'build': u'signing-macosx64-nightly/opt'}

They must get overridden somewhere else

Comment 41

[Kim Moir [:kmoir] ET]

actually the dependencies are fine I just have read too many task graphs today

Comment 42

[Kim Moir [:kmoir] ET]

Attachment: macrepackage.json

latest graph from my try run, Callek I would appreciate feedback on it, I know it's not 100% yet but want to make sure I'm on the correct path

Comment 43

[Justin Wood (:Callek)]

This looks very close, things of note the artifacts should either be a directory listing piece, or a target.dmg entry, not the .zst

We of course need to have the environment piece expanded out (taskref-or-string type thing).

And lastly, the scopes will be needing some love, because of the relengapiproxy features for tooltool (we need the internal scope, at the least).

I do feel this task def is in the right direction.

I think we'll want some tweaking on the index routing, but I'd rather see where its coming from in a real patch before I comment much there.

Comment 44

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-12.patch

updated patches

latest try run is here

https://treeherder.mozilla.org/#/jobs?repo=try&revision=65e323d71db5aa51c76e1644ce68daa4dc460964

Comment 45

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-13.patch

new patches, still have to fix scopes

latest try run 
https://treeherder.mozilla.org/#/jobs?repo=try&revision=61f235ecc75f2c8ff30d3fb0ec87f0e477278e9f

Comment 46

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-14.patch

fixed scopes

I'll also attach json from latest try run to verify that it is okay.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=8ce8845c87861b5fc230ed13bb1c72a4b1172203

Comment 47

[Kim Moir [:kmoir] ET]

Attachment: mac-signing.json

callek does this look better?

Comment 48

[Justin Wood (:Callek)]

Comment on attachment 8850258 [details]
mac-signing.json

Starting to look good,

Current Concerns:

Artifacts: the output name is image.tar.zst instead of target.dmg

Caches: imagebuilder? probably not relevant.  We want at least tooltool's cache as well.

Scopes: Other than appropriate scope for caches, we need `secrets:get:project/taskcluster/gecko/hgfingerprint`
`docker-worker:relengapi-proxy:tooltool.download.public`
in addition to `docker-worker:relengapi-proxy:tooltool.download.internal`

Environment: the value here (in json) is wrong, you have: "SIGNED_INPUT": "https://queue.taskcluster.net/v1/task/{task-reference: <build-signing>}/artifacts/public/build/target.tar.gz",
-- It should have a value of `{task-reference: "https://queue.taskcluster.net/v1/task/<build-signing>/artifacts/public/build/target.tar.gz"}`  needs to actually be a dict.

Image: `"image": "taskcluster/image_builder@`  pretty sure we need the build image.

Routes: I'm not sure we want `index.gecko.v2.try.repackage-nightly.*` as a thing at all, I'd chat with mshal before adding these. (Can omit these routes in pass 1, imo)

Comment 49

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-15.patch

updated patches, will attach json from latest try run

https://treeherder.mozilla.org/#/jobs?repo=try&revision=f55217a147e19a746e8b6f22804109daa9213178

Comment 50

[Kim Moir [:kmoir] ET]

Attachment: mac-repackage.json

Comment 51

[Justin Wood (:Callek)]

Comment on attachment 8850572 [details] [diff] [review]
bug1324052tc-15.patch

Review of attachment 8850572 [details] [diff] [review]:
-----------------------------------------------------------------

::: taskcluster/taskgraph/transforms/repackage.py
@@ +142,5 @@
> +            ['docker-worker:relengapi-proxy:tooltool.download.internal',
> +             'secrets:get:project/taskcluster/gecko/hgfingerprint',
> +             'docker-worker:relengapi-proxy:tooltool.download.public'],
> +            'worker': {'implementation': 'docker-worker',
> +                       'docker-image': docker_image('base-build'),

nope, this yields quay.io results, the image we should be using is the same one as OSX builds from the build kind.

@@ +155,5 @@
> +                                 }],
> +                       'artifacts': [{
> +                                       'type': 'file',
> +                                       'path': '/home/worker/workspace/artifacts/target.dmg',
> +                                       'name': 'public/image.tar.dmg',

name' ... 'target.dmg'  not 'image.tar.dmg'

Comment 52

[Justin Wood (:Callek)]

Comment on attachment 8850572 [details] [diff] [review]
bug1324052tc-15.patch

Noteworthy, iterations of this would be easier (for me) if it was through mozreview, since I could easily see the interdiff of various iterations, rather than trying to skim and see what changed...

Comment 53

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-16.patch

Kims-MacBook-Pro:mozilla-inbound kmoir$ diff ~/bug1324052tc-15.patch   ~/bug1324052tc-16.patch
3c3
< # Parent  ddc1e4a43016fae924da600d07dc37ea5a4e685d
---
> # Parent  c309f93f5c33eaaff2294ef70ab18158d3fbb25d
451c451
< +                       'docker-image': docker_image('base-build'),
---
> +                       'docker-image' : {"in-tree": "desktop-build"},
464c464
< +                                       'name': 'public/image.tar.dmg',
---
> +                                       'name': 'public/image.dmg'

Comment 54

[Kim Moir [:kmoir] ET]

Attachment: mac-repackage2.json

10c10,11
<       "build": "signing-macosx64-nightly/opt"
---
>       "build": "signing-macosx64-nightly/opt",
>       "docker-image": "build-docker-image-desktop-build"
45c46
<         "description": "MacOS X x64 Cross-compile Nightly ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=f55217a147e19a746e8b6f22804109daa9213178)) Signing ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=f55217a147e19a746e8b6f22804109daa9213178)) Repackage ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=f55217a147e19a746e8b6f22804109daa9213178))",
---
>         "description": "MacOS X x64 Cross-compile Nightly ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=5ccfef960e430e48c1d35ea0794396559a60df13)) Signing ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=5ccfef960e430e48c1d35ea0794396559a60df13)) Repackage ([Treeherder push](https://treeherder.mozilla.org/#/jobs?repo=try&revision=5ccfef960e430e48c1d35ea0794396559a60df13))",
48c49
<         "source": "https://hg.mozilla.org/try//file/f55217a147e19a746e8b6f22804109daa9213178/taskcluster/ci/repackage"
---
>         "source": "https://hg.mozilla.org/try//file/5ccfef960e430e48c1d35ea0794396559a60df13/taskcluster/ci/repackage"
52c53
<           "public/image.tar.dmg": {
---
>           "public/image.dmg": {
80c81
<           "GECKO_HEAD_REV": "f55217a147e19a746e8b6f22804109daa9213178",
---
>           "GECKO_HEAD_REV": "5ccfef960e430e48c1d35ea0794396559a60df13",
87c88
<           "MOZ_BUILD_DATE": "20170323160138",
---
>           "MOZ_BUILD_DATE": "20170323175157",
101c102,108
<         "image": "quay.io/mozilla/base-build@sha256:e3fa6fef48461c8a3977f2e328024a1c505b6e78a5f692b964488b45df3a1a2f",
---
>         "image": {
>           "path": "public/image.tar.zst",
>           "taskId": {
>             "task-reference": "<docker-image>"
>           },
>           "type": "task-image"
>         },
106,107c113,114
<         "tc-treeherder.v2.try.f55217a147e19a746e8b6f22804109daa9213178.178244",
<         "tc-treeherder-stage.v2.try.f55217a147e19a746e8b6f22804109daa9213178.178244"
---
>         "tc-treeherder.v2.try.5ccfef960e430e48c1d35ea0794396559a60df13.178268",
>         "tc-treeherder-stage.v2.try.5ccfef960e430e48c1d35ea0794396559a60df13.178268"

Comment 55

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-date.patch

patch for latest on date branch

Comment 56

[Justin Wood (:Callek)]

Comment on attachment 8850615 [details]
mac-repackage2.json

Current concerns:

dependency+SIGNED_INPUT:  the <build-signing> in SIGNED_INPUT won't match the `name` on any of the dependencies. (we have 'build' and 'docker-image' as deps, not 'build-signing') please match them up.

This looks a lot better, thanks!

Comment 57

[Kim Moir [:kmoir] ET]

Attachment: bug1324052tc-17.patch

Thanks for all the reviews Callek. I appreciate it and know I've asked a lot the last few days.

diff between last two patches
430c430
< +        dependencies = {job['depname']: dep_job.label}
---
> +        dependencies = {dep_job.attributes.get('kind'): dep_job.label}

Kims-MacBook-Pro:mozilla-inbound kmoir$ diff ~/mac-repackage2.json ~/mac-repackage3.json
10c10
<       "build": "signing-macosx64-nightly/opt",
---
>       "build-signing": "signing-macosx64-nightly/opt",
https://treeherder.mozilla.org/#/jobs?repo=try&revision=b714019ea37746f99b68c4a99adef086003d882b&selectedJob=86095949


mozreview is on my list of things to start using, I just have a lot of bugzilla muscle memory.

Comment 58

[Kim Moir [:kmoir] ET]

Attachment: mac-repackage3.json

Comment 59

[Kim Moir [:kmoir] ET]

Attachment: updated patch for date

Comment 60

[Aki Sasaki (not active)]

Comment on attachment 8850807 [details] [diff] [review]
bug1324052tc-17.patch

Just looking at the diff you posted, those look like good changes.

Comment 61

[Kim Moir [:kmoir] ET]

kmoir
https://tools.taskcluster.net/task-group-inspector/#/Dmvttt5HRxOwWgvGhuy74g?_k=6efiw7
4:34 PM build is running
4:35 PM  I need to change all the dependencies so that beetmover etc depend on mac repackage not signing
aki
yeah
4:37 PM windows will also need special beetmover etc deps

Comment 62

[Kim Moir [:kmoir] ET]

Attachment: bug1324052-more.patch

I also had to land this patch to add the signing:dmg scope and then add the scope here

 https://tools.taskcluster.net/auth/roles/#project:releng:nightly:level-3:*

Comment 63

[Kim Moir [:kmoir] ET]

The repackage job ran but the link to the dmg artifact says it is missing on the worker

https://tools.taskcluster.net/task-group-inspector/#/Dmvttt5HRxOwWgvGhuy74g/cxi6BhRpRY2f0X4c6cbLNw?_k=f01een

Comment 64

[Aki Sasaki (not active)]

(In reply to Kim Moir [:kmoir] from comment #63)
> The repackage job ran but the link to the dmg artifact says it is missing on
> the worker
> 
> https://tools.taskcluster.net/task-group-inspector/#/Dmvttt5HRxOwWgvGhuy74g/
> cxi6BhRpRY2f0X4c6cbLNw?_k=f01een

From the log:
[task 2017-03-24T21:00:04.731024Z] 21:00:04     INFO - Copy/paste: python2.7 mach --log-no-times repackage --input /home/worker/workspace/inputs/target.tar.gz --output /home/worker/workspace/build/upload/target.dmg

From the task definition:
  "artifacts": {
    "public/image.dmg": {
      "path": "/home/worker/workspace/artifacts/target.dmg",

Looks like you want to change that path to "/home/worker/workspace/build/upload/target.dmg" .  Also, "public/image.dmg" probably wants to be "public/build":

  "artifacts": {
    "public/build": {
      "path": "/home/worker/workspace/build/upload/target.dmg",

Comment 65

[Aki Sasaki (not active)]

https://hg.mozilla.org/projects/date/rev/0d9ede0c99c6f27eb6ff5152f73a4983c54a126b
bug 1324052 - fix repackage artifact upload.

Comment 66

[Aki Sasaki (not active)]

(In reply to Aki Sasaki [:aki] from comment #64)
>   "artifacts": {
>     "public/build": {
>       "path": "/home/worker/workspace/build/upload/target.dmg",

Looks like I have to specify "public/build/target.dmg", probably because this isn't a directory.

Comment 67

[Aki Sasaki (not active)]

https://hg.mozilla.org/projects/date/rev/405e6792178d964fea8b2b3f6a3736579d2e1adb
bug 1324052 - public/build/target.dmg. DONTBUILD

Comment 68

[Kim Moir [:kmoir] ET]

Attachment: bug1324052date.patch

consolidated patches on date

I have to clean them up to remove the logging stuff I was using for debugging as well as fix the schema verification for repackage.

Now, I'll work on fixing the dependencies for beetmover so that they shift to depending on the repackage task for mac, but remain the signing task for other platforms

Comment 69

[Kim Moir [:kmoir] ET]

so the beetmover doesn't can't currently deal with tasks that have two dependencies
and the repackage task  has dependencies like this
 "dependencies": {
      "build-signing": "signing-macosx64-nightly/opt",
      "docker-image": "build-docker-image-desktop-build"
    },

so I think I need to transform the job for macosx in the beetmover transform

Comment 70

[Dustin J. Mitchell [:dustin] (he/him)]

That's the second time beetmover's funny business with dependencies has caused issues.  Maybe it can be reconsidered?  I don't understand the rationale too well, so all I can do is suggest the possibility that maybe the beetmover implementation should be changed.

Comment 71

[Kim Moir [:kmoir] ET]

Attachment: bug1324052beet.patch

so these are preliminary patches to change the dependencies for macosx nightlies so the beetmover task depends on repackage instead of build-signing

they are very hacky. I just changed the beetmover kind to have two dependencies and then changed the beetmover transforms and loader not to load the dependency that's not needed macosx and vice versa

The alternative approach is the add another kind but then you have to add new kinds for all the following tasks because if you have a new beetmover kind for macosx, then for balrog then for checksums etc.

So I'm looking for feedback if this is the right approach or not.

The l10n dependencies are not fixed.

I'll attach the json from a try run in the next patch.

Comment 72

[Kim Moir [:kmoir] ET]

Attachment: beetrepackage.json

beetmover-repackage-macosx64-nightly

from this try run

https://treeherder.mozilla.org/#/jobs?repo=try&revision=cc782df798acab5685b563c2232b36320c33a46a

There are several things wrong with this
for instance 

"task-reference": "<repackage>" in the payload
for the build artifacts

the non-dmg artifacts should be from the build task, yet the dmg needs to be from the signed build

anyways, wip

Comment 73

[Kim Moir [:kmoir] ET]

Mshal, what should be the routes for the repackage task?  Callek mentioned in comment 48 that I should ask you about this.

Comment 74

[Kim Moir [:kmoir] ET]

This bug covered adding macosx nightly builds and the signing and repackage tasks to the taskgraph.

I opened bugs 1352417 (beetmover) and 1352426 (balrog) parts of the taskgraph.  They depend on other beetmover and balrog bugs that need to be implemented before we can add them to the taskgraph.

Comment 75

[Michael Shal [:mshal]]

Do we need to index both the artifacts from the signing task and the repackage task? IMO the signing task is really just intermediate in this case, and we should only index the repackaged task. Callek and I chatted about how to index the final signed artifacts, which I believe are the ones under gecko.v2.mozilla-central.signed-nightly

Comment 76

[Aki Sasaki (not active)]

https://hg.mozilla.org/build/puppet/rev/ca09fef599e812bfbfc9b86b72764781243f95e4
bug 1324052 - bump scriptworker versions to allow for repackage cot. r=callek