Deflake test-api job, currently failing at around 7.6% (44 out of 578
workflows), by ensuring that test nomad agent use a small dedicated port
range that doesn't conflict with the kernel ephemeral range.
The failures are disproportionatly related to port allocation, where a
nomad agent fails to start when the http port is already bound to
another process. The failures are intermitent and aren't specific to any
test in particular. The following is a representative failure:
https://app.circleci.com/pipelines/github/hashicorp/nomad/13995/workflows/6cf6eb38-f93c-46f8-8aa0-f61e62fe7694/jobs/128169
.
Upon investigation, the issue seems to be that the api freeport library
picks a port block within 10,000-14,500, but that overlaps with the
kernel ephemeral range 32,769-60,999! So, freeport may allocate a free
port to the nomad agent, just to be used by another process before the
nomad agent starts!
This happened for example in
https://app.circleci.com/pipelines/github/hashicorp/nomad/14111/workflows/e1fcd7ff-f0e0-4796-8719-f57f510b1ffa/jobs/129684
. `freeport` allocated port 41662 to serf, but `google_accounts`
raced to use it to connect to the CirleCI vm metadata service.
We avoid such races by using a dedicated port range that's disjoint from
the kernel ephemeral port range.
This PR deflakes TestTaskRunner_StatsHook_Periodic tests and adds backoff when the driver closes the channel.
TestTaskRunner_StatsHook_Periodic is currently the most flaky test - failing ~4% of the time (20 out of 486 workflows). A sample failure: https://app.circleci.com/pipelines/github/hashicorp/nomad/14028/workflows/957b674f-cbcc-4228-96d9-1094fdee5b9c/jobs/128563 .
This change has two components:
First, it updates the StatsHook so that it backs off when stats channel is closed. In the context of the test where the mock driver emits a single stats update and closes the channel, the test may make tens of thousands update during the period. In real context, if a driver doesn't implement the stats handler properly or when a task finishes, we may generate way too many Stats queries in a tight loop. Here, the backoff reduces these queries. I've added a failing test that shows 154,458 stats updates within 500ms in https://app.circleci.com/pipelines/github/hashicorp/nomad/14092/workflows/50672445-392d-4661-b19e-e3561ed32746/jobs/129423 .
Second, the test ignores the first stats update after a task exit. Due to the asynchronicity of updates and channel/context use, it's possible that an update is enqueued while the test marks the task as exited, resulting into a spurious update.
This removes modification of ops in methods that UpdateWorkload calls, keeping
them local to UpdateWorkload. It also includes some rewrites of checkRegs for
clarity.
I want to strike a balance here:
- On the one hand there are use cases (raw_exec or Docker only) where
running Nomad clients as an unprivileged user is *preferable.*
- On the other hand running Nomad clients as root is our main and best
tested environment. So I want to leave that a strong recommendation.
There's no reason to give the `client.options` to `plugins` migration
top billing on the client configuration page. Remove and and clarify the
more appropriately placed note down below.
