Update the Consul/Vault build downloader functions so that we include the
current prerelease build (if any) in our E2E compatibility testing we do on each
PR. This will automatically cycle out when the GA build is released, because
that build is "higher" in the sorted set.
* e2e: add tests for exec2 task driver
* e2e: use envoy 1.29.4 because consul
* e2e: add a bridge networking http test for exec driver
* e2e: split up http test so curl always starts after the server
The process by which we tag AMIs with the commit SHA of the Packer directory
isn't documented in this repository, which makes it easy to accidentally build
an AMI that will break nightly E2E.
This change exposes CNI configuration details of a network
namespace as environment variables. This allows a task to use
these value to configure itself; a potential use case is to run
a Raft application binding to IP and Port details configured using
the bridge network mode.
This reverts commit 45b36371a12ffae5b5bfaaeadb08f801fb6bc98d. Now that Vault
1.16.2 has shipped, the E2E test will pick up only a working version.
Closes: https://github.com/hashicorp/nomad/issues/20298
In #20296 we added a Go tool chain to the AMI we use for E2E tests, so that we
can build `consul-cni` for tproxy testing. This is intended to be temporary
until `consul-k8s` 1.4.2 is officially released. But the Go cache from building
`consul-k8s` uses up roughly 1.5GiB of space and the test machines have fairly
small disks. This causes the Nomad clients to aggressively GC client allocations
that stop, which breaks tests that run batch workloads and then read their logs.
Add the `consul-cni` plugin to the Linux AMI for E2E, and add a test case that
covers the transparent proxy feature. Add test assertions to the Connect tests
for upstream reachability
Ref: https://github.com/hashicorp/nomad/pull/20175
The E2E test for periodic dispatch jobs has a `cron` trigger for once a
minute. If the test happens to run at the top of the minute, it's possible for
the forced dispatch to run from the test code, then the periodic timer triggers
and leaves a running child job. This fails the test because it expects only a
single job in the "dead" state.
Make it so that the `cron` expression is implausible to run during our test
window, and migrate the test off the old framework while we're at it.
Vault 1.16.1 has a known issue around the JWT auth configuration that will
prevent this test from ever passing. Skip testing the JWT code path on
1.16.1. Once 1.16.2 ships it will no longer get skipped.
Ref: https://github.com/hashicorp/nomad/issues/20298
Migrate our E2E tests for Connect off the old framework in preparation for
writing E2E tests for transparent proxy and the updated workload identity
workflow. Mark the tests that cover the legacy Consul token submitted workflow.
Ref: https://github.com/hashicorp/nomad/pull/20175
If a E2E cluster is destroyed after a different one has been created, the role
and policy we create in Vault for the cluster will be deleted and Vault-related
tests will fail. Note that before 1.9, we should figure out a way to give HCP
Vault access to the JWKS endpoint and have a different set of policies, but
we'll need to have a role-per-cluster in that case as well.
Fixes: https://github.com/hashicorp/nomad-e2e/issues/138 (internal)
Our `consulcompat` tests exercise both the Workload Identity and legacy Consul
token workflow, but they are limited to running single node tests. The E2E
cluster is network isolated, so using our HCP Consul cluster runs into a
problem validating WI tokens because it can't reach the JWKS endpoint. In real
production environments, you'd solve this with a CNAME pointing to a public IP
pointing to a proxy with a real domain name. But that's logisitcally
impractical for our ephemeral nightly cluster.
Migrate the HCP Consul to a single-node Consul cluster on AWS EC2 alongside our
Nomad cluster. Bootstrap TLS and ACLs in Terraform and ensure all nodes can
reach each other. This will allow us to update our Consul tests so they can use
Workload Identity, in a separate PR.
Ref: #19698
We've been getting a couple of errors from this test on nightly where the
template hasn't rendered by the time we expect it to. I've run some tests
locally and this may be a timing issue introduced by recent code changes to
templates.
Move the start of the timer to after we're guaranteed that we've got a secret
lease TTL started, to eliminate this as a source of flakiness. In my tests this
adds another ~5s to a test that already takes over a minute to run anyways.
* e2e: move rawexec oversub tests into oversubscription e2e test suite
This PR moves two tests for raw_exec and memory oversubscription into
the oversubscription test suite, which has the necessary plumbing to
activate and restore the oversubscription configuration of the scheduler
during the test.
* cr: rename files for better readability
* drivers/raw_exec: enable configuring raw_exec task to have no memory limit
This PR makes it possible to configure a raw_exec task to not have an
upper memory limit, which is how the driver would behave pre-1.7.
This is done by setting memory_max = -1. The cluster (or node pool) must
have memory oversubscription enabled.
* cl: add cl
* drivers/executor: set oom_score_adj for raw_exec
This might not be wholly true since I don't know all configurations of
Nomad, but in our use cases, we run some of our tasks as `raw_exec` for
reasons.
We observed that our tasks were running with `oom_score_adj = -1000`,
which prevents them from being OOM'd. This value is being inherited from
the nomad agent parent process, as configured by systemd.
Similar to #10698, we also were shocked to have this value inherited
down to every child process and believe that we should also set this
value to 0 explicitly.
I have no idea if there are other paths that might leverage this or
other ways that `raw_exec` can manifest, but this is how I was able to
observe and fix in one of our configurations.
We have been running in production our tasks wrapped in a script that
does: `echo 0 > /proc/self/oom_score_adj` to avoid this issue.
* drivers/executor: minor cleanup of setting oom adjustment
* e2e: add test for raw_exec oom adjust score
* e2e: set oom score adjust to -999
* cl: add cl
---------
Co-authored-by: Seth Hoenig <shoenig@duck.com>
* e2e/connect: adds test for namespace policies
* consul: use token namespace when fetching policies
* changelog
* fixup! e2e/connect: adds test for namespace policies
The EBS snapshot operation can take a long time to complete. Recent runs have
shown we sometimes get up to the 10s timeout on the context we're giving the CLI
command. Extend this so that we're not getting spurious timeouts.
Fixes: https://github.com/hashicorp/nomad/issues/19118
* cleanup consul tokens by accessor id
rather than secret id, which has been failing for some time with:
> 404 (Cannot find token to delete)
* expect subset of consul namespaces
the consul test cluster may have namespaces from other unrelated tests
This will dump much of the interesting parts of cluster state, including
available nodes and their status, existing allocations and their status,
and existing evaluations and their status.
When porting the `ConsulTemplate` test, I made a last-minute refactor to the
assertions for waiting on files, and accidentally inverted the test assertion in
the process.
Also, when running `jobs3.Submit` you need to include the `Namespace` option so
that the cleanup function that gets return deletes the job from the correct
namespace. This was causing the namespace cleanup to fail because the job
deletion had failed.
