Sync repo #1143

joshsmith · 2017-10-31T03:02:17Z

What's in this PR?

This is almost good to merge now.

Closes #1111
Closes #1133
Closes #1136
Closes #971

joshsmith · 2017-10-31T03:03:25Z

lib/code_corps/model/github_pull_request.ex

+  end
+  defp add_merged(%Ecto.Changeset{} = changeset) do
+    changeset |> put_change(:merged, false)
+  end


Apparently GitHub does not pass merged along when listing the records.

Can also remove:

additions changed_files comments commits deletions mergeable_state review_comments

joshsmith · 2017-10-31T03:34:27Z

I will mention that it is really, really hard to debug problems happening inside of these Multi fns.

joshsmith · 2017-10-31T04:53:25Z

I fixed an issue where marshall_result was not providing us with the github_issue result for github_pull_requests, but there appears to be a result missing. For our test repo we should be seeing 31 issues created, but it looks like there are only 30, with perhaps a missed validation error.

I wanted to at least get this spike working before worrying too much about error handling. Hard to get it working without it, though.

joshsmith · 2017-10-31T06:13:25Z

I managed to debug it. I'm getting a single task back that's got an {:error, :validating_tasks} that's failing due to the timestamp from GitHub being before the modified_at.

It turns out this is due to a bug: we're not setting the modified_from field to "code_corps" for anything that's being update from the Code Corps app.

Of course, this means my strategy here is probably flawed. We won't get back updated issues/tasks when a sync isn't happening for the first time. We could modify this by returning back the matched issues regardless of whether the sync happens. That would allow my strategy here to continue to work. But it's probably worth stepping back and re-evaluating this approach.

That said, the general outline of how to approach the problem is there as a start.

joshsmith · 2017-10-31T06:23:33Z

I ran a mix ecto.reset and it works all the way through. So this is effectively spiked. But lots of issues obviously remain with how this works.

joshsmith · 2017-10-31T06:54:51Z

lib/code_corps/github/sync/sync.ex

+    tasks = Enum.map(tasks, fn task -> {:ok, task} end)
+    sync_comments(repo, github_issues, tasks)
+    repo
+  end


Inside here we could sync each step, modify the GithubRepo sync state, and then return the repo when done.

begedin · 2017-10-31T14:15:25Z

@joshsmith I managed to spike out a less coupled, fully working GithubRepo sync approach. I've encountered several difficulties while doing this, one of which I'm still trying to find a solution for.

Problems

`/repos/:owner/:name/comments` retrieves repository comments, NOT issue comments

Really, in hindsight, it makes sense, but it completely dodged my attention. The comments retrieved by that endpoint are code comments, not comments from within issues.

I had to implement an API.Issue.comments endpoint and then also implemented a utility endpoint API.Repository.issue_comments which preloads all GithubIssue records for a provided GithubRepo, then calls API.Issue.comments/1 for each of them and aggregates the results.

This makes the request somewhat slow, since it fires off parallel :head requests equal to the issue count on the repo, but it's still faster than doing each of those individually.

To allow for this, I also had to add a function clause to the GithubComment.create_or_update_comment which only requires a payload and associates with an issue matching by Repo.get_by(&1, url: issue_url). This decouples the function and we can likely switch the webhook syncing code to using it as well. More importantly, it allows for a simple pipe during the repo sync process.

i believe we need a `sync_state` on the `ProjectGithubRepo` as well

Once a GithubRepo is synced, and it receiving webhooks, there is no need to sync it again. When a project is connected to it, the only thing that's left to do is to create necessary Task and Comment records on that Project.

I added a virtual sync_state field to the ProjectGithubRepo for now, just so I could spike things out, but we should a proper field. I'm thinking the states are

~w(unsynced syncing_tasks errored_syncinc_tasks syncinc_comments errored_syncinc_comments)

our current sync flow does not allow creating `Task` or `Comment` records separate from `GithubIssue` and `GithubComment` respectively.

This is the issue that I'm stuck with.

The major part of the reason we introduced GithubIssue and GithubComment as separate entities was to decouple this process and allow easier updating when a single issue needs to sync as a task across multiple projects, etc.

However, due to the way we match the user, there is no way for us to actually separate the two. The user is matched from the initial payload, but no user information is stored on the GithubIssue/GithubComment for later use.

I considered several potential solutions for this

Make fetching payloads a process separate from the main syncing flow

Basically, we would sync a project_github_repo, along the lines of

with {:ok, pr_payloads} <- repo |> fetch_pull_requests,
        {:ok, issue_payloads} <- repo |> fetch_issues,
        {:ok, comment_payloads} <- repo |> fetch_comments,
        {:ok, %GithubRepo{sync_state: "receiving_webhooks"} = repo} <- repo |> sync_repo(pr_payloads, issue_payloads, comment_payloads),
        {:ok, %ProjectGithubRepo{sync_state: "synced"} = project_github_repo} <- project_github_repo |> sync_project_github_repo(issue_payloads, comment_payloads) 
do
  {:ok, project_githubrepo}
end

Then, we would sync a project by syncing each of its project_github_repo children.

Really, though, this is a bandaid fix, since it does not resolve the actual problem of too tight coupling and just allows us to implement the sync process sooner.

Try to make it work like webhooks work, keep it coupled and perform the sync of `Task/GithubIssue` as well as `Comment/GithubComment` in unison

I don't often say this, but I'm strongly against this option.

The whole process is complicated and hard to understand as is, and this will only make it worse.

Add `github_user_id` field to `GithubIssue` and `GithubComment`

This would be a temporary fix, but it would possibly allow us to defer the full solution into a separate PR.

Basically, we store a github user id field with the records, allowing us to match later. However, this does not provide us with enough information for the case when the author of the comment/issue does not have a codecorps accoutn linked to github, so that case would be unhandled.

I don't favor this solution, but wanted to mention it for the sake of argument.

Introduce `GithubUser`

As part of this PR we

introduce a GithubUser record
as we sync comments and issues, a GithubUser record is found or created alongside each of them

The record basically holds the fields we've added to User

field :avatar_url, :string
field :email, :string
field :github_id, :integer
field :username, :string
field :type, :string, default: "User" # or "Bot"

Once we start syncing tasks and comments, we match a GithubUser with a User (find or create) and associate appropriately.

As part of subsequent PRs we can then

Switch the webhook syncing code to use GithubUser to. We could do it here as well, but we don't have to.
Potentially decouple our webhook syncing code, consolidate further, move out of multis, etc.
Potentially clean up our User model a bit
???

begedin · 2017-10-31T15:23:29Z

We could speed up comment fetching with Repository.issue_comments by storing comment_count on the GithubIssue (there is a "comments": comment_count key in the github payload) and then only sync github issues with comment count greater than 0.

begedin · 2017-10-31T16:17:49Z

joshsmith · 2017-10-31T23:51:06Z

For adding GithubUser we'll need to:

add references to User, GithubIssue, GithubPullRequest, GithubComment
add all these to models
add model tests
add all these to views and view tests

joshsmith · 2017-11-01T01:48:53Z

This is going to need some significant cleanup of changesets and the way changes are propagated to the various GithubX models during sync.

joshsmith · 2017-11-01T05:28:42Z

This is mostly working, but on a first pass through a project, it seems to not create tasks in the syncing_tasks step, moves onto syncing_comments, and then hits a nil Task that it can't match on:

** (WithClauseError) no with clause matching: nil
    (code_corps) lib/code_corps/github/sync/comment/comment/comment.ex:63: CodeCorps.GitHub.Sync.Comment.Comment.find_or_create_comment/2
    (elixir) lib/enum.ex:1255: Enum."-map/2-lists^map/1-0-"/2
    (code_corps) lib/code_corps/github/sync/comment/comment/comment.ex:58: CodeCorps.GitHub.Sync.Comment.Comment.sync_project_github_repo/1
    (code_corps) lib/code_corps/github/sync/sync.ex:163: CodeCorps.GitHub.Sync.sync_project_github_repo/1

joshsmith · 2017-11-01T06:36:38Z

I wonder if it is worth adding in a synced_at timestamp to the GithubRepo and ProjectGithubRepo.

joshsmith · 2017-11-01T07:58:14Z

My first goals tomorrow:

begedin

This is pretty much implemented the way I would continue off.

I got a suggested issue for adapter behavior, but that's about it.

begedin · 2017-11-01T08:22:31Z

lib/code_corps/github/adapters/issue.ex

+    github_issue
+    |> Map.from_struct
+    |> MapTransformer.transform_inverse(@issue_mapping)
+  end


The naming makes sense, but considering we were switching already, I'm thinking maybe we want to write a behavior module and document what an adapter needs to look like and what the naming convention is.

I'm just imagining where, in a couple of weeks, I write a new one and end up using a new scheme because I wasn't 100% what the old one is.

joshsmith · 2017-11-02T00:16:08Z

This is missing tests for the GitHub.Sync that hit the error states for repo and project_repo sync.

joshsmith

Some notes on things we can still improve here.

joshsmith · 2017-11-02T00:30:53Z

lib/code_corps/accounts/changesets.ex

+    |> unique_github_constraint()
+  end
+
+  def update_with_github_user_changeset(struct, %{} = params) do


Missing spec and docs.

joshsmith · 2017-11-02T00:31:03Z

lib/code_corps/github/adapters/comment.ex

    {:url, ["url"]}
  ]

+  def to_comment_attrs(%GithubComment{} = github_comment) do


Missing spec and docs.

joshsmith · 2017-11-02T00:31:09Z

lib/code_corps/github/adapters/issue.ex

  end

+  def to_issue_attrs(%GithubIssue{} = github_issue) do
+    github_issue


Missing spec and docs.

joshsmith · 2017-11-02T00:31:37Z

lib/code_corps/github/sync/comment/comment/comment.ex

    |> ResultAggregator.aggregate
  end

+  def sync_project_github_repo(%ProjectGithubRepo{github_repo: %GithubRepo{} = _} = project_github_repo) do


Missing spec and doc.

joshsmith · 2017-11-02T00:32:05Z

lib/code_corps/github/sync/issue/task/task.ex

    |> ResultAggregator.aggregate
  end

+  def sync_project_github_repo(%ProjectGithubRepo{github_repo: %GithubRepo{} = _} = project_github_repo) do


Missing spec and doc.

joshsmith · 2017-11-02T00:32:39Z

lib/code_corps/github/sync/sync.ex

+    end
+  end
+
+  defp put_sync_opt(opts, key, map) do


Missing spec and doc.

joshsmith · 2017-11-02T00:32:44Z

lib/code_corps/github/sync/sync.ex

+  end
+
+  @spec sync_repo(GithubRepo.t) :: {:ok, GithubRepo.t}
+  def sync_repo(%GithubRepo{} = repo) do


Missing doc.

joshsmith · 2017-11-02T00:32:50Z

lib/code_corps/github/sync/sync.ex

+    end
+  end
+
+  def sync_project_github_repo(%ProjectGithubRepo{} = project_github_repo) do


Missing spec and doc.

joshsmith · 2017-11-02T00:33:41Z

lib/code_corps/validators/time_validator.ex

+      nil -> changeset
+      _ -> do_validate_time_after(changeset, field, previous_time, current_time)
+    end
+  end


Needs updated unit tests.

joshsmith · 2017-11-02T00:33:53Z

lib/code_corps_web/controllers/project_github_repo_controller.ex

         {:ok, :authorized} <- current_user |> Policy.authorize(:create, %ProjectGithubRepo{}, params),
         {:ok, %ProjectGithubRepo{} = project_github_repo} <- create_project_repo_changeset(params) |> Repo.insert do

+      TaskSupervisor.start_child(:background_processor, fn -> CodeCorps.GitHub.Sync.sync_project_github_repo(project_github_repo) end)


This is not tested anywhere right now.

begedin

A couple of ideas I have, plus a potential change that needs to be reverted, but other than that, this looks great!

begedin · 2017-11-02T14:55:51Z

lib/code_corps/github/sync/sync.ex

+
+    with {:ok, project_github_repo} <- project_github_repo |> mark_project_repo("syncing_github_repo"),
+         {:ok, %GithubRepo{sync_state: "receiving_webhooks"}} <- repo |> sync_repo(),
+         project_github_repo <- Repo.get(ProjectGithubRepo, project_github_repo.id) |> preload_project_github_repo(),


I'd actually prefer something like

preloads <- [:project, github_repo: [:github_app_installation, [github_issues: [:github_comments, :github_user]]]], project_github_repo <- project_github_repo |> Repo.preload(preloads)

I think it's actually easier to follow. Either that, or renaming preload_project_gihub_repo to load_associations.

Seems somewhat more explicit to me.

begedin · 2017-11-02T15:00:31Z

lib/code_corps/github/sync/sync.ex

+         {:ok, %GithubRepo{sync_state: "receiving_webhooks"}} <- repo |> sync_repo(),
+         project_github_repo <- Repo.get(ProjectGithubRepo, project_github_repo.id) |> preload_project_github_repo(),
+         {:ok, project_github_repo} <- project_github_repo |> mark_project_repo("syncing_tasks"),
+         {:ok, _tasks} <- project_github_repo |> Sync.Issue.Task.sync_project_github_repo() |> sync_step(:sync_tasks),


Trying to read Sync.Issue.Task.sync_project_github_repo makes me think we may have nested too deep.

project_github_repo |> Sync.TaskSyncer.sync

Looks a bit clearer, but I'm still not happy with it. However, the we could alias Sync.TaskSyncer and just call

project_github_repo |> TaskSyncer.sync

What do you think about flattening out our Sync namespace that way? Nesting Task and GithubIssue under Issue together seems like we did just for the sake of them being semantically related, but logically, we probably want the two processes as decoupled as possible.

sync/sync.ex sync/task_syncer.ex sync/github_issue_syncer.ex

etc. looks nicer to me.

For now, as part of this PR, I'm thinking we

alias CodeCorps.GitHub.Sync.Issue.Task, as: TaskSyncer

and do the same for other modules. What do you think?

begedin · 2017-11-02T15:01:07Z

lib/code_corps/github/sync/sync.ex

+  defp marshall_result({:ok, %{github_pull_request: _, github_issue: _}} = result), do: result
  defp marshall_result({:ok, %{github_pull_request: pull_request}}), do: {:ok, pull_request}
+  defp marshall_result({:ok, %{github_issue: _, tasks: _}} = result), do: result
+  defp marshall_result({:ok, %{github_issue: github_issue}}), do: {:ok, github_issue}


Are these still needed? I think we didn't touch transactions in this PR

joshsmith assigned begedin Oct 31, 2017

joshsmith commented Oct 31, 2017

View reviewed changes

begedin reviewed Nov 1, 2017

View reviewed changes

joshsmith force-pushed the 1111-sync-comments branch from 35b302b to 76acde5 Compare November 2, 2017 00:25

joshsmith commented Nov 2, 2017

View reviewed changes

begedin reviewed Nov 2, 2017

View reviewed changes

joshsmith force-pushed the 1111-sync-comments branch from e0dab0a to 6141cc3 Compare November 4, 2017 02:19

Add syncing process for GitHub repos

33cf7b3

joshsmith force-pushed the 1111-sync-comments branch from 18dd750 to 33cf7b3 Compare November 4, 2017 23:07

joshsmith merged commit 886434b into develop Nov 4, 2017

joshsmith deleted the 1111-sync-comments branch November 4, 2017 23:15

Uh oh!

Sync repo #1143

Sync repo #1143

Uh oh!

Conversation

joshsmith commented Oct 31, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What's in this PR?

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joshsmith commented Oct 31, 2017

Uh oh!

joshsmith commented Oct 31, 2017

Uh oh!

joshsmith commented Oct 31, 2017

Uh oh!

joshsmith commented Oct 31, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

begedin commented Oct 31, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problems

/repos/:owner/:name/comments retrieves repository comments, NOT issue comments

i believe we need a sync_state on the ProjectGithubRepo as well

our current sync flow does not allow creating Task or Comment records separate from GithubIssue and GithubComment respectively.

Make fetching payloads a process separate from the main syncing flow

Try to make it work like webhooks work, keep it coupled and perform the sync of Task/GithubIssue as well as Comment/GithubComment in unison

Add github_user_id field to GithubIssue and GithubComment

Introduce GithubUser

As part of this PR we

As part of subsequent PRs we can then

Uh oh!

begedin commented Oct 31, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

begedin commented Oct 31, 2017 • edited by joshsmith Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joshsmith commented Oct 31, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joshsmith commented Nov 1, 2017

Uh oh!

joshsmith commented Nov 1, 2017

Uh oh!

joshsmith commented Nov 1, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joshsmith commented Nov 1, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

begedin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

joshsmith commented Nov 2, 2017

Uh oh!

joshsmith left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

joshsmith commented Oct 31, 2017 •

edited

Loading

begedin commented Oct 31, 2017 •

edited

Loading

`/repos/:owner/:name/comments` retrieves repository comments, NOT issue comments

i believe we need a `sync_state` on the `ProjectGithubRepo` as well

our current sync flow does not allow creating `Task` or `Comment` records separate from `GithubIssue` and `GithubComment` respectively.

Try to make it work like webhooks work, keep it coupled and perform the sync of `Task/GithubIssue` as well as `Comment/GithubComment` in unison

Add `github_user_id` field to `GithubIssue` and `GithubComment`

Introduce `GithubUser`

begedin commented Oct 31, 2017 •

edited

Loading

begedin commented Oct 31, 2017 •

edited by joshsmith

Loading

joshsmith commented Oct 31, 2017 •

edited

Loading

joshsmith commented Nov 1, 2017 •

edited

Loading

joshsmith commented Nov 1, 2017 •

edited

Loading