This seed downloads 2.1 GB worth of repositories. Google can afford the
bandwidth, but if a person using the GDK is on a metered connection,
that's not so great.
Also the GDK test suite runs this seed, so every CI run for that project
had to download those as well. Needlessly wasteful.
There were two cases that could be problematic:
1. Because sometimes AuthorizedProjectsWorker would be scheduled in a
transaction it was possible for a job to run/complete before a
COMMIT; resulting in it either producing an error, or producing no
new data.
2. When scheduling jobs the code would not wait until completion. This
could lead to a user creating a project and then immediately trying
to push to it. Usually this will work fine, but given enough load it
might take a few seconds before a user has access.
The first one is problematic, the second one is mostly just annoying
(but annoying enough to warrant a solution).
This commit changes two things to deal with this:
1. Sidekiq scheduling now takes places after a COMMIT, this is ensured
by scheduling using Rails' after_commit hook instead of doing so in
an arbitrary method.
2. When scheduling jobs the calling thread now waits for all jobs to
complete.
Solution 2 requires tracking of job completions. Sidekiq provides a way
to find a job by its ID, but this involves scanning over the entire
queue; something that is very in-efficient for large queues. As such a
more efficient solution is necessary. There are two main Gems that can
do this in a more efficient manner:
* sidekiq-status
* sidekiq_status
No, this is not a joke. Both Gems do a similar thing (but slightly
different), and the only difference in their name is a dash vs an
underscore. Both Gems however provide far more than just checking if a
job has been completed, and both have their problems. sidekiq-status
does not appear to be actively maintained, with the last release being
in 2015. It also has some issues during testing as API calls are not
stubbed in any way. sidekiq_status on the other hand does not appear to
be very popular, and introduces a similar amount of code.
Because of this I opted to write a simple home grown solution. After
all, all we need is storing a job ID somewhere so we can efficiently
look it up; we don't need extra web UIs (as provided by sidekiq-status)
or complex APIs to update progress, etc.
This is where Gitlab::SidekiqStatus comes in handy. This namespace
contains some code used for tracking, removing, and looking up job IDs;
all without having to scan over an entire queue. Data is removed
explicitly, but also expires automatically just in case.
Using this API we can now schedule jobs in a fork-join like manner: we
schedule the jobs in Sidekiq, process them in parallel, then wait for
completion. By using Sidekiq we can leverage all the benefits such as
being able to scale across multiple cores and hosts, retrying failed
jobs, etc.
The one downside is that we need to make sure we can deal with
unexpected increases in job processing timings. To deal with this the
class Gitlab::JobWaiter (used for waiting for jobs to complete) will
only wait a number of seconds (30 by default). Once this timeout is
reached it will simply return.
For GitLab.com almost all AuthorizedProjectWorker jobs complete in
seconds, only very rarely do we spike to job timings of around a minute.
These in turn seem to be the result of external factors (e.g. deploys),
in which case a user is most likely not able to use the system anyway.
In short, this new solution should ensure that jobs are processed
properly and that in almost all cases a user has access to their
resources whenever they need to have access.
When I proposed using serializable transactions I was hoping we would be
able to refresh data of individual users concurrently. Unfortunately
upon closer inspection it was revealed this was not the case. This could
result in a lot of queries failing due to serialization errors,
overloading the database in the process (given enough workers trying to
update the target table).
To work around this we're now using a Redis lease that is cancelled upon
completion. This ensures we can update the data of different users
concurrently without overloading the database.
The code will try to obtain the lease until it succeeds, waiting at
least 1 second between retries. This is necessary as we may otherwise
end up _not_ updating the data which is not an option.
1. Change multiple updates to a single `update_all`
2. Use cascading deletes
3. Extract an average function for the database median.
4. Move database median to `lib/gitlab/database`
5. Use `delete_all` instead of `destroy_all`
6. Minor refactoring
- The normal seed creates all the data for cycle analytics the "right"
way. It creates issues, merge requests, commits, branches,
deployments, etc. This is good, but too slow for perf testing.
Generating a 1000 sets of records this way takes more than an hour.
- When the `CYCLE_ANALYTICS_POPULATE_METRICS_DIRECTLY` environment
variable is passed in, the seed only creates issues and merge
requests. It then adds the `metrics` for each issue and
merge request directly, to save time.
- The seed now takes about 4 minutes to run for 1000 sets of records.
1. These changes bring down page load time for 100 issues from more than
a minute to about 1.5 seconds.
2. This entire commit is composed of these types of performance
enhancements:
- Cache relevant data in `IssueMetrics` wherever possible.
- Cache relevant data in `MergeRequestMetrics` wherever possible.
- Preload metrics
3. Given these improvements, we now only need to make 4 SQL calls:
- Load all issues
- Load all merge requests
- Load all metrics for the issues
- Load all metrics for the merge requests
4. A list of all the data points that are now being pre-calculated:
a. The first time an issue is mentioned in a commit
- In `GitPushService`, find all issues mentioned by the given commit
using `ReferenceExtractor`. Set the `first_mentioned_in_commit_at`
flag for each of them.
- There seems to be a (pre-existing) bug here - files (and
therefore commits) created using the Web CI don't have
cross-references created, and issues are not closed even when
the commit title is "Fixes #xx".
b. The first time a merge request is deployed to production
When a `Deployment` is created, find all merge requests that
were merged in before the deployment, and set the
`first_deployed_to_production_at` flag for each of them.
c. The start / end time for a merge request pipeline
Hook into the `Pipeline` state machine. When the `status` moves to
`running`, find the merge requests whose tip commit matches the
pipeline, and record the `latest_build_started_at` time for each
of them. When the `status` moves to `success`, record the
`latest_build_finished_at` time.
d. The merge requests that close an issue
- This was a big cause of the performance problems we were having
with Cycle Analytics. We need to use `ReferenceExtractor` to make
this calculation, which is slow when we have to run it on a large
number of merge requests.
- When a merge request is created, updated, or refreshed, find the
issues it closes, and create an instance of
`MergeRequestsClosingIssues`, which acts as a join model between
merge requests and issues.
- If a `MergeRequestsClosingIssues` instance links a merge request
and an issue, that issue closes that merge request.
5. The `Queries` module was changed into a class, so we can cache the
results of `issues` and `merge_requests_closing_issues` across
various cycle analytics stages.
6. The code added in this commit is untested. Tests will be added in the
next commit.
- The fixture generates data for every stage in the cycle analytics
dashboard. Once this fixture has run, you shouldn't be seeing any
"<not enough data>" messages for cycle analytics.
- This is probably not necessary for every fixture run, so it might be
moved behind an env var in the future.
