This PR introduces support for using Nomad on systems with cgroups v2 [1]
enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux
distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems
for Nomad users.
Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer,
but not so for managing cpuset cgroups. Before, Nomad has been making use of
a feature in v1 where a PID could be a member of more than one cgroup. In v2
this is no longer possible, and so the logic around computing cpuset values
must be modified. When Nomad detects v2, it manages cpuset values in-process,
rather than making use of cgroup heirarchy inheritence via shared/reserved
parents.
Nomad will only activate the v2 logic when it detects cgroups2 is mounted at
/sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2
mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to
use the v1 logic, and should operate as before. Systems that do not support
cgroups v2 are also not affected.
When v2 is activated, Nomad will create a parent called nomad.slice (unless
otherwise configured in Client conifg), and create cgroups for tasks using
naming convention <allocID>-<task>.scope. These follow the naming convention
set by systemd and also used by Docker when cgroups v2 is detected.
Client nodes now export a new fingerprint attribute, unique.cgroups.version
which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by
Nomad.
The new cpuset management strategy fixes#11705, where docker tasks that
spawned processes on startup would "leak". In cgroups v2, the PIDs are
started in the cgroup they will always live in, and thus the cause of
the leak is eliminated.
[1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.htmlCloses#11289Fixes#11705#11773#11933
* Use unix:// prefix for CSI_ENDPOINT variable by default
* Some plugins have strict validation over the format of the
`CSI_ENDPOINT` variable, and unfortunately not all plugins
agree. Allow the user to override the `CSI_ENDPOINT` to workaround
those cases.
* Update all demos and tests with CSI_ENDPOINT
This commit performs refactoring to pull out common service
registration objects into a new `client/serviceregistration`
package. This new package will form the base point for all
client specific service registration functionality.
The Consul specific implementation is not moved as it also
includes non-service registration implementations; this reduces
the blast radius of the changes as well.
The Prestart hook for task runner hooks doesn't get called when we
restore a task, because the task is already running. The Postrun hook
for CSI plugin supervisors needs the socket path to have been
populated so that the client has a valid path.
When the alloc runner claims a volume, an allocation for a previous
version of the job may still have the volume claimed because it's
still shutting down. In this case we'll receive an error from the
server. Retry this error until we succeed or until a very long timeout
expires, to give operators a chance to recover broken plugins.
Make the alloc runner hook tolerant of temporary RPC failures.
The dynamic plugin registry assumes that plugins are singletons, which
matches the behavior of other Nomad plugins. But because dynamic
plugins like CSI are implemented by allocations, we need to handle the
possibility of multiple allocations for a given plugin type + ID, as
well as behaviors around interleaved allocation starts and stops.
Update the data structure for the dynamic registry so that more recent
allocations take over as the instance manager singleton, but we still
preserve the previous running allocations so that restores work
without racing.
Multiple allocations can run on a client for the same plugin, even if
only during updates. Provide each plugin task a unique path for the
control socket so that the tasks don't interfere with each other.
In PR #11892 we updated the `csi_hook` to unmount the volume locally
via the CSI node RPCs before releasing the claim from the server. The
timer for this hook was initialized with the retry time, forcing us to
wait 1s before making the first unmount RPC calls.
Use the new helper for timers to ensure we clean up the timer nicely.
This PR modifies the Consul CLI arguments used to bootstrap envoy for
Connect sidecars to make use of '-proxy-id' instead of '-sidecar-for'.
Nomad registers the sidecar service, so we know what ID it has. The
'-sidecar-for' was intended for use when you only know the name of the
service for which the sidecar is being created.
The improvement here is that using '-proxy-id' does not require an underlying
request for listing Consul services. This will make make the interaction
between Nomad and Consul more efficient.
Closes#10452
When Consul Connect just works, it's wonderful. When it doesn't work it
can be exceeding difficult to debug: operators have to check task
events, Nomad logs, Consul logs, Consul APIs, and even then critical
information is missing.
Using Consul to generate a bootstrap config for Envoy is notoriously
difficult. Nomad doesn't even log stderr, so operators are left trying
to piece together what went wrong.
This patch attempts to provide *maximal* context which unfortunately
includes secrets. **Secrets are always restricted to the secrets/
directory.** This makes debugging a little harder, but allows operators
to know exactly what operation Nomad was trying to perform.
What's added:
- stderr is sent to alloc/logs/envoy_bootstrap.stderr.0
- the CLI is written to secrets/.envoy_bootstrap.cmd
- the environment is written to secrets/.envoy_bootstrap.env as JSON
Accessing this information is unfortunately awkward:
```
nomad alloc exec -task connect-proxy-count-countdash b36a cat secrets/.envoy_bootstrap.env
nomad alloc exec -task connect-proxy-count-countdash b36a cat secrets/.envoy_bootstrap.cmd
nomad alloc fs b36a alloc/logs/envoy_bootstrap.stderr.0
```
The above assumes an alloc id that starts with `b36a` and a Connect
sidecar proxy for a service named `count-countdash`.
If the alloc is unable to start successfully, the debugging files are
only accessible from the host filesystem.
Nomad communicates with CSI plugin tasks via gRPC. The plugin
supervisor hook uses this to ping the plugin for health checks which
it emits as task events. After the first successful health check the
plugin supervisor registers the plugin in the client's dynamic plugin
registry, which in turn creates a CSI plugin manager instance that has
its own gRPC client for fingerprinting the plugin and sending mount
requests.
If the plugin manager instance fails to connect to the plugin on its
first attempt, it exits. The plugin supervisor hook is unaware that
connection failed so long as its own pings continue to work. A
transient failure during plugin startup may mislead the plugin
supervisor hook into thinking the plugin is up (so there's no need to
restart the allocation) but no fingerprinter is started.
* Refactors the gRPC client to connect on first use. This provides the
plugin manager instance the ability to retry the gRPC client
connection until success.
* Add a 30s timeout to the plugin supervisor so that we don't poll
forever waiting for a plugin that will never come back up.
Minor improvements:
* The plugin supervisor hook creates a new gRPC client for every probe
and then throws it away. Instead, reuse the client as we do for the
plugin manager.
* The gRPC client constructor has a 1 second timeout. Clarify that this
timeout applies to the connection and not the rest of the client
lifetime.
The CSI specification says:
> The CO SHALL provide the listen-address for the Plugin by way of the
`CSI_ENDPOINT` environment variable.
Note that plugins without filesystem isolation won't have the plugin
dir bind-mounted to their alloc dir, but we can provide a path to the
socket anyways.
Refactor to use opts struct for plugin supervisor hook config.
The parameter list for configuring the plugin supervisor hook has
grown enough where is makes sense to use an options struct similiar to
many of the other task runner hooks (ex. template).
In certain task lifecycles the taskrunner service deregister call
could be called three times for a task that is exiting. Whilst
each hook caller of deregister has its own purpose, we should try
and ensure it is only called once during the shutdown lifecycle of
a task.
This change therefore tracks when deregister has been called, so
that subsequent calls are noop. In the event the task is
restarting, the deregister value is reset to ensure proper
operation.
go-getter creates a circular dependency between a Client and Getter,
which means each is inherently thread-unsafe if you try to re-use
on or the other.
This PR fixes Nomad to no longer make use of the default Getter objects
provided by the go-getter package. Nomad must create a new Client object
on every artifact download, as the Client object controls the Src and Dst
among other things. When Caling Client.Get, the Getter modifies its own
Client reference, creating the circular reference and race condition.
We can still achieve most of the desired connection caching behavior by
re-using a shared HTTP client with transport pooling enabled.
This PR replaces use of time.After with a safe helper function
that creates a time.Timer to use instead. The new function returns
both a time.Timer and a Stop function that the caller must handle.
Unlike time.NewTimer, the helper function does not panic if the duration
set is <= 0.
Previously we copied this library by hand to avoid vendor-ing a bunch of
files related to minimock. Now that we no longer vendor, just import the
library normally.
Also we might use more of the library for handling `time.After` uses,
for which this library provides a Context-based solution.
When an allocation stops, the `csi_hook` makes an unpublish RPC to the
servers to unpublish via the CSI RPCs: first to the node plugins and
then the controller plugins. The controller RPCs must happen after the
node RPCs so that the node has had a chance to unmount the volume
before the controller tries to detach the associated device.
But the client has local access to the node plugins and can
independently determine if it's safe to send unpublish RPC to those
plugins. This will allow the server to treat the node plugin as
abandoned if a client is disconnected and `stop_on_client_disconnect`
is set. This will let the server try to send unpublish RPCs to the
controller plugins, under the assumption that the client will be
trying to unmount the volume on its end first.
Note that the CSI `NodeUnpublishVolume`/`NodeUnstageVolume` RPCs can
return ignorable errors in the case where the volume has already been
unmounted from the node. Handle all other errors by retrying until we
get success so as to give operators the opportunity to reschedule a
failed node plugin (ex. in the case where they accidentally drained a
node without `-ignore-system`). Fan-out the work for each volume into
its own goroutine so that we can release a subset of volumes if only
one is stuck.
When an allocation stops, the `csi_hook` makes an unpublish RPC to the
servers to unpublish via the CSI RPCs: first to the node plugins and
then the controller plugins. The controller RPCs must happen after the
node RPCs so that the node has had a chance to unmount the volume
before the controller tries to detach the associated device.
But the client has local access to the node plugins and can
independently determine if it's safe to send unpublish RPC to those
plugins. This will allow the server to treat the node plugin as
abandoned if a client is disconnected and `stop_on_client_disconnect`
is set. This will let the server try to send unpublish RPCs to the
controller plugins, under the assumption that the client will be
trying to unmount the volume on its end first.
Note that the CSI `NodeUnpublishVolume`/`NodeUnstageVolume` RPCs can
return ignorable errors in the case where the volume has already been
unmounted from the node. Handle all other errors by retrying until we
get success so as to give operators the opportunity to reschedule a
failed node plugin (ex. in the case where they accidentally drained a
node without `-ignore-system`). Fan-out the work for each volume into
its own goroutine so that we can release a subset of volumes if only
one is stuck.
Templates in nomad jobs make use of the vault token defined in
the vault stanza when issuing credentials like client certificates.
When using change_mode "noop" in the vault stanza, consul-template
is not informed in case a vault token is re-issued (which can
happen from time to time for various reasons, as described
in https://www.nomadproject.io/docs/job-specification/vault).
As a result, consul-template will keep using the old vault token
to renew credentials and - once the token expired - stop renewing
credentials. The symptom of this problem is a vault_token
file that is newer than the issued credential (e.g., TLS certificate)
in a job's /secrets directory.
This change corrects this, so that h.updater.updatedVaultToken(token)
is called, which will inform stakeholders about the new
token and make sure, the new token is used by consul-template.
Example job template fragment:
vault {
policies = ["nomad-job-policy"]
change_mode = "noop"
}
template {
data = <<-EOH
{{ with secret "pki_int/issue/nomad-job"
"common_name=myjob.service.consul" "ttl=90m"
"alt_names=localhost" "ip_sans=127.0.0.1"}}
{{ .Data.certificate }}
{{ .Data.private_key }}
{{ .Data.issuing_ca }}
{{ end }}
EOH
destination = "${NOMAD_SECRETS_DIR}/myjob.crt"
change_mode = "noop"
}
This fix does not alter the meaning of the three change modes of vault
- "noop" - Take no action
- "restart" - Restart the job
- "signal" - send a signal to the task
as the switch statement following line 232 contains the necessary
logic.
It is assumed that "take no action" was never meant to mean "don't tell
consul-template about the new vault token".
Successfully tested in a staging cluster consisting of multiple
nomad client nodes.
This PR exposes the following existing`consul-template` configuration options to Nomad jobspec authors in the `{job.group.task.template}` stanza.
- `wait`
It also exposes the following`consul-template` configuration to Nomad operators in the `{client.template}` stanza.
- `max_stale`
- `block_query_wait`
- `consul_retry`
- `vault_retry`
- `wait`
Finally, it adds the following new Nomad-specific configuration to the `{client.template}` stanza that allows Operators to set bounds on what `jobspec` authors configure.
- `wait_bounds`
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Co-authored-by: Michael Schurter <mschurter@hashicorp.com>
The task runner prestart hooks take a `joincontext` so they have the
option to exit early if either of two contexts are canceled: from
killing the task or client shutdown. Some tasks exit without being
shutdown from the server, so neither of the joined contexts ever gets
canceled and we leak the `joincontext` (48 bytes) and its internal
goroutine. This primarily impacts batch jobs and any task that fails
or completes early such as non-sidecar prestart lifecycle tasks.
Cancel the `joincontext` after the prestart call exits to fix the
leak.
Some operators use very long group/task `shutdown_delay` settings to
safely drain network connections to their workloads after service
deregistration. But during incident response, they may want to cause
that drain to be skipped so they can quickly shed load.
Provide a `-no-shutdown-delay` flag on the `nomad alloc stop` and
`nomad job stop` commands that bypasses the delay. This sets a new
desired transition state on the affected allocations that the
allocation/task runner will identify during pre-kill on the client.
Note (as documented here) that using this flag will almost always
result in failed inbound network connections for workloads as the
tasks will exit before clients receive updated service discovery
information and won't be gracefully drained.
Fixes#2522
Skip embedding client.alloc_dir when building chroot. If a user
configures a Nomad client agent so that the chroot_env will embed the
client.alloc_dir, Nomad will happily infinitely recurse while building
the chroot until something horrible happens. The best case scenario is
the filesystem's path length limit is hit. The worst case scenario is
disk space is exhausted.
A bad agent configuration will look something like this:
```hcl
data_dir = "/tmp/nomad-badagent"
client {
enabled = true
chroot_env {
# Note that the source matches the data_dir
"/tmp/nomad-badagent" = "/ohno"
# ...
}
}
```
Note that `/ohno/client` (the state_dir) will still be created but not
`/ohno/alloc` (the alloc_dir).
While I cannot think of a good reason why someone would want to embed
Nomad's client (and possibly server) directories in chroots, there
should be no cause for harm. chroots are only built when Nomad runs as
root, and Nomad disables running exec jobs as root by default. Therefore
even if client state is copied into chroots, it will be inaccessible to
tasks.
Skipping the `data_dir` and `{client,server}.state_dir` is possible, but
this PR attempts to implement the minimum viable solution to reduce risk
of unintended side effects or bugs.
When running tests as root in a vm without the fix, the following error
occurs:
```
=== RUN TestAllocDir_SkipAllocDir
alloc_dir_test.go:520:
Error Trace: alloc_dir_test.go:520
Error: Received unexpected error:
Couldn't create destination file /tmp/TestAllocDir_SkipAllocDir1457747331/001/nomad/test/testtask/nomad/test/testtask/.../nomad/test/testtask/secrets/.nomad-mount: open /tmp/TestAllocDir_SkipAllocDir1457747331/001/nomad/test/.../testtask/secrets/.nomad-mount: file name too long
Test: TestAllocDir_SkipAllocDir
--- FAIL: TestAllocDir_SkipAllocDir (22.76s)
```
Also removed unused Copy methods on AllocDir and TaskDir structs.
Thanks to @eveld for not letting me forget about this!
Add a new hostname string parameter to the network block which
allows operators to specify the hostname of the network namespace.
Changing this causes a destructive update to the allocation and it
is omitted if empty from API responses. This parameter also supports
interpolation.
In order to have a hostname passed as a configuration param when
creating an allocation network, the CreateNetwork func of the
DriverNetworkManager interface needs to be updated. In order to
minimize the disruption of future changes, rather than add another
string func arg, the function now accepts a request struct along with
the allocID param. The struct has the hostname as a field.
The in-tree implementations of DriverNetworkManager.CreateNetwork
have been modified to account for the function signature change.
In updating for the change, the enhancement of adding hostnames to
network namespaces has also been added to the Docker driver, whilst
the default Linux manager does not current implement it.
When creating a TCP proxy bridge for Connect tasks, we are at the
mercy of either end for managing the connection state. For long
lived gRPC connections the proxy could reasonably expect to stay
open until the context was cancelled. For the HTTP connections used
by connect native tasks, we experience connection disconnects.
The proxy gets recreated as needed on follow up requests, however
we also emit a WARN log when the connection is broken. This PR
lowers the WARN to a TRACE, because these disconnects are to be
expected.
Ideally we would be able to proxy at the HTTP layer, however Consul
or the connect native task could be configured to expect mTLS, preventing
Nomad from MiTM the requests.
We also can't mange the proxy lifecycle more intelligently, because
we have no control over the HTTP client or server and how they wish
to manage connection state.
What we have now works, it's just noisy.
Fixes#10933
This PR implements a new "System Batch" scheduler type. Jobs can
make use of this new scheduler by setting their type to 'sysbatch'.
Like the name implies, sysbatch can be thought of as a hybrid between
system and batch jobs - it is for running short lived jobs intended to
run on every compatible node in the cluster.
As with batch jobs, sysbatch jobs can also be periodic and/or parameterized
dispatch jobs. A sysbatch job is considered complete when it has been run
on all compatible nodes until reaching a terminal state (success or failed
on retries).
Feasibility and preemption are governed the same as with system jobs. In
this PR, the update stanza is not yet supported. The update stanza is sill
limited in functionality for the underlying system scheduler, and is
not useful yet for sysbatch jobs. Further work in #4740 will improve
support for the update stanza and deployments.
Closes#2527
This PR fixes a bug where the underlying Envoy process of a Connect gateway
would consume a full core of CPU if there is more than one sidecar or gateway
in a group. The utilization was being caused by Consul injecting an envoy_ready_listener
on 127.0.0.1:8443, of which only one of the Envoys would be able to bind to.
The others would spin in a hot loop trying to bind the listener.
As a workaround, we now specify -address during the Envoy bootstrap config
step, which is how Consul maps this ready listener. Because there is already
the envoy_admin_listener, and we need to continue supporting running gateways
in host networking mode, and in those case we want to use the same port
value coming from the service.port field, we now bind the admin listener to
the 127.0.0.2 loop-back interface, and the ready listener takes 127.0.0.1.
This shouldn't make a difference in the 99.999% use case where envoy is
being run in its official docker container. Advanced users can reference
${NOMAD_ENVOY_ADMIN_ADDR_<service>} (as they 'ought to) if needed,
as well as the new variable ${NOMAD_ENVOY_READY_ADDR_<service>} for the
envoy_ready_listener.
