Data can be pushed to GitHub without creating a Commit record.

[alloy]

There’s a bug that makes it that a POST to github actually succeeds, but no Commit record is ever created. Looking at PodVersion#push! it would seem that the this might occur because response.success? is returning false when it actually did succeeded.

Because the user thinks the push failed, they might try again and run into an issue with the UPDATE variant of GitHub’s contents API (opposed to CREATE) requiring a SHA. This returns a 422 to us and we return a 500 to the user.

An example, which I have since fixed manually:

irb(main):001:0> Pod::TrunkApp::Pod.find(:name => 'DB5-Live')
I, [2014-05-23T21:25:24.179119 #2]  INFO -- : (0.004418s) SELECT * FROM "pods" WHERE ("name" = 'DB5-Live') LIMIT 1
=> #<Pod::TrunkApp::Pod @values={:id=>5066, :name=>"DB5-Live", :created_at=>2014-05-23 17:16:04 UTC, :updated_at=>nil}>
irb(main):002:0> p = _
=> #<Pod::TrunkApp::Pod @values={:id=>5066, :name=>"DB5-Live", :created_at=>2014-05-23 17:16:04 UTC, :updated_at=>nil}>
irb(main):003:0> p.versions
I, [2014-05-23T21:25:30.309944 #2]  INFO -- : (0.003187s) SELECT * FROM "pod_versions" WHERE ("pod_versions"."pod_id" = 5066)
=> [#<Pod::TrunkApp::PodVersion @values={:id=>15510, :name=>"0.1.4", :created_at=>2014-05-23 17:16:04 UTC, :updated_at=>nil, :pod_id=>5066}>]
irb(main):004:0> p.versions.first
=> #<Pod::TrunkApp::PodVersion @values={:id=>15510, :name=>"0.1.4", :created_at=>2014-05-23 17:16:04 UTC, :updated_at=>nil, :pod_id=>5066}>
irb(main):005:0> v = _
=> #<Pod::TrunkApp::PodVersion @values={:id=>15510, :name=>"0.1.4", :created_at=>2014-05-23 17:16:04 UTC, :updated_at=>nil, :pod_id=>5066}>
irb(main):006:0> v.commits
I, [2014-05-23T21:26:33.659066 #2]  INFO -- : (0.002274s) SELECT * FROM "commits" WHERE ("commits"."pod_version_id" = 15510) ORDER BY ("updated_at", "created_at") ASC
=> []

[alloy]

A second thing to consider is that the GitHub commit web hook POST, which is meant to check and fix the integrity of the DB, did not do its job.

[floere]

Good point. So there was a new version pushed and the GH hook did not add the commit for it?

We should definitely add a test case for this scenario!

[alloy]

Yep, I still don’t understand how that could happen, though.

It might make sense to start by adding a search field to /manage/log_messages so that we can search for all the pushes that have had this issue. E.g. search for status code ‘422’.