Closed Bug 1001224 Opened 10 years ago Closed 10 years ago

Updates to gaia-node-modules are burning TBPL

Categories

(Firefox OS Graveyard :: Gaia, defect)

x86
macOS
defect
Not set
normal

Tracking

(Not tracked)

RESOLVED FIXED
2.0 S1 (9may)

People

(Reporter: kgrandon, Assigned: kgrandon)

References

Details

(Whiteboard: [p=3],[systemsfe])

Attachments

(5 files)

From my dev-gaia email:

We have had to do a few backouts recently due to gaia-node-modules updates burning TBPL. I think there is something wrong with sockit-to-me rebuilding on TBPL. I am not entirely sure what's going on yet, so if anyone has any ideas it would be greatly appreciated. We are seeing errors like:

https://tbpl.mozilla.org/php/getParsedLog.php?id=38375867&tree=B2g-Inbound
https://tbpl.mozilla.org/php/getParsedLog.php?id=38432781&tree=B2g-Inbound

In the meantime, please avoid updating package.json or gaia_node_modules.revision without pushing to gaia-try first.

We think this may be related to the fact that there are two versions of sockit-to-me in the modules dependency tree. We should probably just be using 0.2.0.
Blocks: 998776
Blocks: 997045
Do we really need the dependency twice? It seems that we have circular once.

Definitely having the same version of sockit-to-me everywhere would help.
First of all, thanks for looking into this issue.


I found that some binaries of sockit-to-me would be updated when we we try to do "npm install marionette-client", as shown in the following changes,
https://github.com/mozilla-b2g/gaia-node-modules/commit/b7d10840775e9699e60881f662794121b5800cc5

This binaries updating would happen even when I installed marionette-client 1.1.5 (current in-use version in Gaia), and 1.1.5 should be using sockit-to-me 0.2 right now.
Not sure why sockit-to-me got changed while no obvious version update.
Duh. This shouldn't happen: it is definitely wrong.
Guess I'll take this.
Assignee: nobody → kgrandon
Status: NEW → ASSIGNED
I think this is mostly manifest bumps, so going to go with R=me unless there is anything different that needs to happen.

First step is to land a marionette-plugin-forms update.
After looking further, it seems that I never updated the binaries for sockit-to-me in the npm module.
Attached file Gaia pull request
(In reply to Hubert Figuiere [:hub] from comment #7)
> After looking further, it seems that I never updated the binaries for
> sockit-to-me in the npm module.

Hmm, if we are going to keep using this extension we really need an automated solution for it =/
Anyway, I'm testing this change on try to see what it does, but it sounds like we may need to do another round of package bumps.

https://tbpl.mozilla.org/?tree=Try&rev=88e0f40d6bab
Depends on: 1001533
On try again with the latest sockit-to-me: https://tbpl.mozilla.org/?tree=Try&rev=99304ae95f8b
Whiteboard: [p=1],[systemsfe]
Target Milestone: --- → 2.0 S1 (9may)
Hub - I can't seem to figure out if we got these binaries in properly. I'm now seeing the following error, any chance you could look into it?

https://tbpl.mozilla.org/php/getParsedLog.php?id=38657462&tree=Try

Thanks!
Flags: needinfo?(hub)
Also pinging Aus here so he is aware and may have some insights.
Flags: needinfo?(aus)
Blocks: 986206
Did we compile both 64 and 32bit linux binaries? We need both for tbpl. Is there also a reason why we can't have tbpl compile the module during install? I know that travis needs it pre-compiled but, maybe we can case it so that tbpl always gets a fresh binary.

I think Gareth worked on making this function as expected on Travis so I'm CC'ing him too.
Flags: needinfo?(aus) → needinfo?(gaye)
The remaining problem is a sockit-to-me one, so unassigning from myself.

Aus/Hub - we *really* need a fix for this so we can iterate on modules. If we don't have one by EoD we're probably going to need to move forward with reverting to 0.1.8 so we can unblock people.
Assignee: kgrandon → nobody
(In reply to Ghislain Aus Lacroix [:aus] from comment #14)
> Did we compile both 64 and 32bit linux binaries? We need both for tbpl. 

Yes I did.

> Is
> there also a reason why we can't have tbpl compile the module during
> install? I know that travis needs it pre-compiled but, maybe we can case it
> so that tbpl always gets a fresh binary.
> 
> I think Gareth worked on making this function as expected on Travis so I'm
> CC'ing him too.

It worked before fine. The issue rose because I didn't realize I didn't rebuild the binaries, and then somebody replaced them by mistake.
Flags: needinfo?(hub)
(In reply to Kevin Grandon :kgrandon from comment #15)
> The remaining problem is a sockit-to-me one, so unassigning from myself.
> 
> Aus/Hub - we *really* need a fix for this so we can iterate on modules. If
> we don't have one by EoD we're probably going to need to move forward with
> reverting to 0.1.8 so we can unblock people.

I'll what I can do. It worked for a while fine so.... If only the whole thing was not that fragile *sigh*
Here the problem kevin. In gaia-node-module:

./node_modules/marionette-client/node_modules/sockit-to-me/build/Release/sockit.node:            Mach-O 64-bit x86_64 bundle

Because this should NEVER have been checked in.

Will fix it.
Attached file Breakage fix.
With this it should not be a problem.
Attachment #8414803 - Flags: review?(kgrandon)
(In reply to Hubert Figuiere [:hub] from comment #18)
> Here the problem kevin. In gaia-node-module:
> 
> ./node_modules/marionette-client/node_modules/sockit-to-me/build/Release/
> sockit.node:            Mach-O 64-bit x86_64 bundle
> 
> Because this should NEVER have been checked in.
> 
> Will fix it.

Thanks Hub. Let's go with this for now and test it on try. If files are getting checked in that should not, then we need to update some gitignore files or documentation.
Flags: needinfo?(gaye)
Attachment #8414803 - Flags: review?(kgrandon) → review+
Latest gaia-node-modules updates are on gaia-try now: https://tbpl.mozilla.org/?tree=Try&rev=434d0ba4e3de
Attached file Gitignore update
Hub - I think this could fix inadvertent updates in the future? Look good to you?
Attachment #8414986 - Flags: review?(hub)
Comment on attachment 8414986 [details] [review]
Gitignore update

I had a similar .gitignore ready for a PR.
Attachment #8414986 - Flags: review?(hub) → review+
Try is green, so I think we are going to be good to land here.
Assignee: nobody → kgrandon
Whiteboard: [p=1],[systemsfe] → [p=3],[systemsfe]
Let's see if this sticks: https://github.com/mozilla-b2g/gaia/commit/68e4ae3bc6af77091635e0b72d831748e6c2a620

Thanks for the work on this Hub!
Status: ASSIGNED → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: