This changeset adds the node pool as a label anywhere we're already emitting
labels with additional information such as node class or ID about the client.
The allocrunner sends several updates to the server during the early lifecycle
of an allocation and its tasks. Clients batch-up allocation updates every 200ms,
but experiments like the C2M challenge has shown that even with this batching,
servers can be overwhelmed with client updates during high volume
deployments. Benchmarking done in #9451 has shown that client updates can easily
represent ~70% of all Nomad Raft traffic.
Each allocation sends many updates during its lifetime, but only those that
change the `ClientStatus` field are critical for progressing a deployment or
kicking off a reschedule to recover from failures.
Add a priority to the client allocation sync and update the `syncTicker`
receiver so that we only send an update if there's a high priority update
waiting, or on every 5th tick. This means when there are no high priority
updates, the client will send updates at most every 1s instead of
200ms. Benchmarks have shown this can reduce overall Raft traffic by 10%, as
well as reduce client-to-server RPC traffic.
This changeset also switches from a channel-based collection of updates to a
shared buffer, so as to split batching from sending and prevent backpressure
onto the allocrunner when the RPC is slow. This doesn't have a major performance
benefit in the benchmarks but makes the implementation of the prioritized update
simpler.
Fixes: #9451
Tools like `nomad-nodesim` are unable to implement a minimal implementation of
an allocrunner so that we can test the client communication without having to
lug around the entire allocrunner/taskrunner code base. The allocrunner was
implemented with an interface specifically for this purpose, but there were
circular imports that made it challenging to use in practice.
Move the AllocRunner interface into an inner package and provide a factory
function type. Provide a minimal test that exercises the new function so that
consumers have some idea of what the minimum implementation required is.
When client nodes are restarted, all allocations that have been scheduled on the
node have their modify index updated, including terminal allocations. There are
several contributing factors:
* The `allocSync` method that updates the servers isn't gated on first contact
with the servers. This means that if a server updates the desired state while
the client is down, the `allocSync` races with the `Node.ClientGetAlloc`
RPC. This will typically result in the client updating the server with "running"
and then immediately thereafter "complete".
* The `allocSync` method unconditionally sends the `Node.UpdateAlloc` RPC even
if it's possible to assert that the server has definitely seen the client
state. The allocrunner may queue-up updates even if we gate sending them. So
then we end up with a race between the allocrunner updating its internal state
to overwrite the previous update and `allocSync` sending the bogus or duplicate
update.
This changeset adds tracking of server-acknowledged state to the
allocrunner. This state gets checked in the `allocSync` before adding the update
to the batch, and updated when `Node.UpdateAlloc` returns successfully. To
implement this we need to be able to equality-check the updates against the last
acknowledged state. We also need to add the last acknowledged state to the
client state DB, otherwise we'd drop unacknowledged updates across restarts.
The client restart test has been expanded to cover a variety of allocation
states, including allocs stopped before shutdown, allocs stopped by the server
while the client is down, and allocs that have been completely GC'd on the
server while the client is down. I've also bench tested scenarios where the task
workload is killed while the client is down, resulting in a failed restore.
Fixes#16381
Adds a new configuration to clients to optionally allow them to drain their
workloads on shutdown. The client sends the `Node.UpdateDrain` RPC targeting
itself and then monitors the drain state as seen by the server until the drain
is complete or the deadline expires. If it loses connection with the server, it
will monitor local client status instead to ensure allocations are stopped
before exiting.
In #16217 we switched clients using Consul discovery to the `Status.Members`
endpoint for getting the list of servers so that we're using the correct
address. This endpoint has an authorization gate, so this fails if the anonymous
policy doesn't have `node:read`. We also can't check the `AuthToken` for the
request for the client secret, because the client hasn't yet registered so the
server doesn't have anything to compare against.
Instead of hitting the `Status.Peers` or `Status.Members` RPC endpoint, use the
Consul response directly. Update the `registerNode` method to handle the list of
servers we get back in the response; if we get a "no servers" or "no path to
region" response we'll kick off discovery again and retry immediately rather
than waiting 15s.
* Update ioutil deprecated library references to os and io respectively
* Deal with the errors produced.
Add error handling to filEntry info
Add error handling to info
Nomad servers can advertise independent IP addresses for `serf` and
`rpc`. Somewhat unexpectedly, the `serf` address is also used for both Serf and
server-to-server RPC communication (including Raft RPC). The address advertised
for `rpc` is only used for client-to-server RPC. This split was introduced
intentionally in Nomad 0.8.
When clients are using Consul discovery for connecting to servers, they get an
initial discovery set from Consul and use the correct `rpc` tag in Consul to get
a list of adddresses for servers. The client then makes a `Status.Peers` RPC to
get the list of those servers that are raft peers. But this endpoint is shared
between servers and clients, and provides the address used for Raft.
Most of the time this is harmless because servers will bind on 0.0.0.0 anyways.,
But in topologies where servers are on a private network and clients are on
separate subnets (or even public subnets), clients will make initial contact
with the server to get the list of peers but then populate their local server
set with unreachable addresses.
Cluster administrators can work around this problem by using `server_join` with
specific IP addresses (or DNS names), because the `Node.UpdateStatus` endpoint
returns the correct set of RPC addresses when updating the node. So once a
client has registered, it will get the correct set of RPC addresses.
This changeset updates the client logic to query `Status.Members` instead of
`Status.Peers`, and then extract the correctly advertised address and port from
the response body.
Fixes#14617
Dynamic Node Metadata allows Nomad users, and their jobs, to update Node metadata through an API. Currently Node metadata is only reloaded when a Client agent is restarted.
Includes new UI for editing metadata as well.
---------
Co-authored-by: Phil Renaud <phil.renaud@hashicorp.com>
* client: sandbox go-getter subprocess with landlock
This PR re-implements the getter package for artifact downloads as a subprocess.
Key changes include
On all platforms, run getter as a child process of the Nomad agent.
On Linux platforms running as root, run the child process as the nobody user.
On supporting Linux kernels, uses landlock for filesystem isolation (via go-landlock).
On all platforms, restrict environment variables of the child process to a static set.
notably TMP/TEMP now points within the allocation's task directory
kernel.landlock attribute is fingerprinted (version number or unavailable)
These changes make Nomad client more resilient against a faulty go-getter implementation that may panic, and more secure against bad actors attempting to use artifact downloads as a privilege escalation vector.
Adds new e2e/artifact suite for ensuring artifact downloading works.
TODO: Windows git test (need to modify the image, etc... followup PR)
* landlock: fixup items from cr
* cr: fixup tests and go.mod file
* scheduler: allow updates after alloc reconnects
When an allocation reconnects to a cluster the scheduler needs to run
special logic to handle the reconnection, check if a replacement was
create and stop one of them.
If the allocation kept running while the node was disconnected, it will
be reconnected with `ClientStatus: running` and the node will have
`Status: ready`. This combination is the same as the normal steady state
of allocation, where everything is running as expected.
In order to differentiate between the two states (an allocation that is
reconnecting and one that is just running) the scheduler needs an extra
piece of state.
The current implementation uses the presence of a
`TaskClientReconnected` task event to detect when the allocation has
reconnected and thus must go through the reconnection process. But this
event remains even after the allocation is reconnected, causing all
future evals to consider the allocation as still reconnecting.
This commit changes the reconnect logic to use an `AllocState` to
register when the allocation was reconnected. This provides the
following benefits:
- Only a limited number of task states are kept, and they are used for
many other events. It's possible that, upon reconnecting, several
actions are triggered that could cause the `TaskClientReconnected`
event to be dropped.
- Task events are set by clients and so their timestamps are subject
to time skew from servers. This prevents using time to determine if
an allocation reconnected after a disconnect event.
- Disconnect events are already stored as `AllocState` and so storing
reconnects there as well makes it the only source of information
required.
With the new logic, the reconnection logic is only triggered if the
last `AllocState` is a disconnect event, meaning that the allocation has
not been reconnected yet. After the reconnection is handled, the new
`ClientStatus` is store in `AllocState` allowing future evals to skip
the reconnection logic.
* scheduler: prevent spurious placement on reconnect
When a client reconnects it makes two independent RPC calls:
- `Node.UpdateStatus` to heartbeat and set its status as `ready`.
- `Node.UpdateAlloc` to update the status of its allocations.
These two calls can happen in any order, and in case the allocations are
updated before a heartbeat it causes the state to be the same as a node
being disconnected: the node status will still be `disconnected` while
the allocation `ClientStatus` is set to `running`.
The current implementation did not handle this order of events properly,
and the scheduler would create an unnecessary placement since it
considered the allocation was being disconnected. This extra allocation
would then be quickly stopped by the heartbeat eval.
This commit adds a new code path to handle this order of events. If the
node is `disconnected` and the allocation `ClientStatus` is `running`
the scheduler will check if the allocation is actually reconnecting
using its `AllocState` events.
* rpc: only allow alloc updates from `ready` nodes
Clients interact with servers using three main RPC methods:
- `Node.GetAllocs` reads allocation data from the server and writes it
to the client.
- `Node.UpdateAlloc` reads allocation from from the client and writes
them to the server.
- `Node.UpdateStatus` writes the client status to the server and is
used as the heartbeat mechanism.
These three methods are called periodically by the clients and are done
so independently from each other, meaning that there can't be any
assumptions in their ordering.
This can generate scenarios that are hard to reason about and to code
for. For example, when a client misses too many heartbeats it will be
considered `down` or `disconnected` and the allocations it was running
are set to `lost` or `unknown`.
When connectivity is restored the to rest of the cluster, the natural
mental model is to think that the client will heartbeat first and then
update its allocations status into the servers.
But since there's no inherit order in these calls the reverse is just as
possible: the client updates the alloc status and then heartbeats. This
results in a state where allocs are, for example, `running` while the
client is still `disconnected`.
This commit adds a new verification to the `Node.UpdateAlloc` method to
reject updates from nodes that are not `ready`, forcing clients to
heartbeat first. Since this check is done server-side there is no need
to coordinate operations client-side: they can continue sending these
requests independently and alloc update will succeed after the heartbeat
is done.
* chagelog: add entry for #15068
* code review
* client: skip terminal allocations on reconnect
When the client reconnects with the server it synchronizes the state of
its allocations by sending data using the `Node.UpdateAlloc` RPC and
fetching data using the `Node.GetClientAllocs` RPC.
If the data fetch happens before the data write, `unknown` allocations
will still be in this state and would trigger the
`allocRunner.Reconnect` flow.
But when the server `DesiredStatus` for the allocation is `stop` the
client should not reconnect the allocation.
* apply more code review changes
* scheduler: persist changes to reconnected allocs
Reconnected allocs have a new AllocState entry that must be persisted by
the plan applier.
* rpc: read node ID from allocs in UpdateAlloc
The AllocUpdateRequest struct is used in three disjoint use cases:
1. Stripped allocs from clients Node.UpdateAlloc RPC using the Allocs,
and WriteRequest fields
2. Raft log message using the Allocs, Evals, and WriteRequest fields
3. Plan updates using the AllocsStopped, AllocsUpdated, and Job fields
Adding a new field that would only be used in one these cases (1) made
things more confusing and error prone. While in theory an
AllocUpdateRequest could send allocations from different nodes, in
practice this never actually happens since only clients call this method
with their own allocations.
* scheduler: remove logic to handle exceptional case
This condition could only be hit if, somehow, the allocation status was
set to "running" while the client was "unknown". This was addressed by
enforcing an order in "Node.UpdateStatus" and "Node.UpdateAlloc" RPC
calls, so this scenario is not expected to happen.
Adding unnecessary code to the scheduler makes it harder to read and
reason about it.
* more code review
* remove another unused test
The artifact getter uses the go-getter library to fetch files from
different sources. Any bug in this library that results in a panic can
cause the entire Nomad client to crash due to a single file download
attempt.
This change aims to guard against this types of crashes by recovering
from panics when the getter attempts to download an artifact. The
resulting panic is converted to an error that is stored as a task event
for operator visibility and the panic stack trace is logged to the
client's log.
* cleanup: refactor MapStringStringSliceValueSet to be cleaner
* cleanup: replace SliceStringToSet with actual set
* cleanup: replace SliceStringSubset with real set
* cleanup: replace SliceStringContains with slices.Contains
* cleanup: remove unused function SliceStringHasPrefix
* cleanup: fixup StringHasPrefixInSlice doc string
* cleanup: refactor SliceSetDisjoint to use real set
* cleanup: replace CompareSliceSetString with SliceSetEq
* cleanup: replace CompareMapStringString with maps.Equal
* cleanup: replace CopyMapStringString with CopyMap
* cleanup: replace CopyMapStringInterface with CopyMap
* cleanup: fixup more CopyMapStringString and CopyMapStringInt
* cleanup: replace CopySliceString with slices.Clone
* cleanup: remove unused CopySliceInt
* cleanup: refactor CopyMapStringSliceString to be generic as CopyMapOfSlice
* cleanup: replace CopyMap with maps.Clone
* cleanup: run go mod tidy
This PR implements support for check_restart for checks registered
in the Nomad service provider.
Unlike Consul, Nomad service checks never report a "warning" status,
and so the check_restart.ignore_warnings configuration is not valid
for Nomad service checks.
Log lines which include an error should use the full term "error"
as the context key. This provides consistency across the codebase
and avoids a Go style which operators might not be aware of.
This PR refactors the code path in Client startup for setting up the cpuset
cgroup manager (non-linux systems not affected).
Before, there was a logic bug where we would try to read the cpuset.cpus.effective
cgroup interface file before ensuring nomad's parent cgroup existed. Therefor that
file would not exist, and the list of useable cpus would be empty. Tasks started
thereafter would not have a value set for their cpuset.cpus.
The refactoring fixes some less than ideal coding style. Instead we now bootstrap
each cpuset manager type (v1/v2) within its own constructor. If something goes
awry during bootstrap (e.g. cgroups not enabled), the constructor returns the
noop implementation and logs a warning.
Fixes#14229
* allocrunner: handle lifecycle when all tasks die
When all tasks die the Coordinator must transition to its terminal
state, coordinatorStatePoststop, to unblock poststop tasks. Since this
could happen at any time (for example, a prestart task dies), all states
must be able to transition to this terminal state.
* allocrunner: implement different alloc restarts
Add a new alloc restart mode where all tasks are restarted, even if they
have already exited. Also unifies the alloc restart logic to use the
implementation that restarts tasks concurrently and ignores
ErrTaskNotRunning errors since those are expected when restarting the
allocation.
* allocrunner: allow tasks to run again
Prevent the task runner Run() method from exiting to allow a dead task
to run again. When the task runner is signaled to restart, the function
will jump back to the MAIN loop and run it again.
The task runner determines if a task needs to run again based on two new
task events that were added to differentiate between a request to
restart a specific task, the tasks that are currently running, or all
tasks that have already run.
* api/cli: add support for all tasks alloc restart
Implement the new -all-tasks alloc restart CLI flag and its API
counterpar, AllTasks. The client endpoint calls the appropriate restart
method from the allocrunner depending on the restart parameters used.
* test: fix tasklifecycle Coordinator test
* allocrunner: kill taskrunners if all tasks are dead
When all non-poststop tasks are dead we need to kill the taskrunners so
we don't leak their goroutines, which are blocked in the alloc restart
loop. This also ensures the allocrunner exits on its own.
* taskrunner: fix tests that waited on WaitCh
Now that "dead" tasks may run again, the taskrunner Run() method will
not return when the task finishes running, so tests must wait for the
task state to be "dead" instead of using the WaitCh, since it won't be
closed until the taskrunner is killed.
* tests: add tests for all tasks alloc restart
* changelog: add entry for #14127
* taskrunner: fix restore logic.
The first implementation of the task runner restore process relied on
server data (`tr.Alloc().TerminalStatus()`) which may not be available
to the client at the time of restore.
It also had the incorrect code path. When restoring a dead task the
driver handle always needs to be clear cleanly using `clearDriverHandle`
otherwise, after exiting the MAIN loop, the task may be killed by
`tr.handleKill`.
The fix is to store the state of the Run() loop in the task runner local
client state: if the task runner ever exits this loop cleanly (not with
a shutdown) it will never be able to run again. So if the Run() loops
starts with this local state flag set, it must exit early.
This local state flag is also being checked on task restart requests. If
the task is "dead" and its Run() loop is not active it will never be
able to run again.
* address code review requests
* apply more code review changes
* taskrunner: add different Restart modes
Using the task event to differentiate between the allocrunner restart
methods proved to be confusing for developers to understand how it all
worked.
So instead of relying on the event type, this commit separated the logic
of restarting an taskRunner into two methods:
- `Restart` will retain the current behaviour and only will only restart
the task if it's currently running.
- `ForceRestart` is the new method where a `dead` task is allowed to
restart if its `Run()` method is still active. Callers will need to
restart the allocRunner taskCoordinator to make sure it will allow the
task to run again.
* minor fixes
In #14139 this code was changed to use the original copy of the config,
but Config.AllocDir is updated in the `Client.init()` method for dev
agents.
This uses the latest version of the alloc dir (which cannot change
further at runtime without a client restart which would reinitialize
the stats collector as well).
Before this change, Client had 2 copies of the config object: config and configCopy. There was no guidance around which to use where (other than configCopy's comment to pass it to alloc runners), both are shared among goroutines and mutated in data racy ways. At least at one point I think the idea was to have `config` be mutable and then grab a lock to overwrite `configCopy`'s pointer atomically. This would have allowed alloc runners to read their config copies in data race safe ways, but this isn't how the current implementation worked.
This change takes the following approach to safely handling configs in the client:
1. `Client.config` is the only copy of the config and all access must go through the `Client.configLock` mutex
2. Since the mutex *only protects the config pointer itself and not fields inside the Config struct:* all config mutation must be done on a *copy* of the config, and then Client's config pointer is overwritten while the mutex is acquired. Alloc runners and other goroutines with the old config pointer will not see config updates.
3. Deep copying is implemented on the Config struct to satisfy the previous approach. The TLS Keyloader is an exception because it has its own internal locking to support mutating in place. An unfortunate complication but one I couldn't find a way to untangle in a timely fashion.
4. To facilitate deep copying I made an *internally backward incompatible API change:* our `helper/funcs` used to turn containers (slices and maps) with 0 elements into nils. This probably saves a few memory allocations but makes it very easy to cause panics. Since my new config handling approach uses more copying, it became very difficult to ensure all code that used containers on configs could handle nils properly. Since this code has caused panics in the past, I fixed it: nil containers are copied as nil, but 0-element containers properly return a new 0-element container. No more "downgrading to nil!"
This PR adds support for specifying checks in services registered to
the built-in nomad service provider.
Currently only HTTP and TCP checks are supported, though more types
could be added later.
Fix numerous go-getter security issues:
- Add timeouts to http, git, and hg operations to prevent DoS
- Add size limit to http to prevent resource exhaustion
- Disable following symlinks in both artifacts and `job run`
- Stop performing initial HEAD request to avoid file corruption on
retries and DoS opportunities.
**Approach**
Since Nomad has no ability to differentiate a DoS-via-large-artifact vs
a legitimate workload, all of the new limits are configurable at the
client agent level.
The max size of HTTP downloads is also exposed as a node attribute so
that if some workloads have large artifacts they can specify a high
limit in their jobspecs.
In the future all of this plumbing could be extended to enable/disable
specific getters or artifact downloading entirely on a per-node basis.
This PR adds support for the raw_exec driver on systems with only cgroups v2.
The raw exec driver is able to use cgroups to manage processes. This happens
only on Linux, when exec_driver is enabled, and the no_cgroups option is not
set. The driver uses the freezer controller to freeze processes of a task,
issue a sigkill, then unfreeze. Previously the implementation assumed cgroups
v1, and now it also supports cgroups v2.
There is a bit of refactoring in this PR, but the fundamental design remains
the same.
Closes#12351#12348
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
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.
Nomad inherited protocol version numbering configuration from Consul and
Serf, but unlike those projects Nomad has never used it. Nomad's
`protocol_version` has always been `1`.
While the code is effectively unused and therefore poses no runtime
risks to leave, I felt like removing it was best because:
1. Nomad's RPC subsystem has been able to evolve extensively without
needing to increment the version number.
2. Nomad's HTTP API has evolved extensively without increment
`API{Major,Minor}Version`. If we want to version the HTTP API in the
future, I doubt this is the mechanism we would choose.
3. The presence of the `server.protocol_version` configuration
parameter is confusing since `server.raft_protocol` *is* an important
parameter for operators to consider. Even more confusing is that
there is a distinct Serf protocol version which is included in `nomad
server members` output under the heading `Protocol`. `raft_protocol`
is the *only* protocol version relevant to Nomad developers and
operators. The other protocol versions are either deadcode or have
never changed (Serf).
4. If we were to need to version the RPC, HTTP API, or Serf protocols, I
don't think these configuration parameters and variables are the best
choice. If we come to that point we should choose a versioning scheme
based on the use case and modern best practices -- not this 6+ year
old dead code.
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 `consul.client_auto_join` configuration block tells the Nomad
client whether to use Consul service discovery to find Nomad
servers. By default it is set to `true`, but contrary to the
documentation it was only respected during the initial client
registration. If a client missed a heartbeat, failed a
`Node.UpdateStatus` RPC, or if there was no Nomad leader, the client
would fallback to Consul even if `client_auto_join` was set to
`false`. This changeset returns early from the client's trigger for
Consul discovery if the `client_auto_join` field is set to `false`.
