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.
If an allocation is slow to stop because of `kill_timeout` or `shutdown_delay`,
the node drain is marked as complete prematurely, even though drain monitoring
will continue to report allocation migrations. This impacts the UI or API
clients that monitor node draining to shut down nodes.
This changeset updates the behavior to wait until the client status of all
drained allocs are terminal before marking the node as done draining.
* func: add namespace support for list deployment
* func: add wildcard to namespace filter for deployments
* Update deployment_endpoint.go
* style: use must instead of require or asseert
* style: rename paginator to avoid clash with import
* style: add changelog entry
* fix: add missing parameter for upsert jobs
msgpackrpc codec handles are specific to a connection and cannot be shared
between goroutines; this can cause corrupted decoding. Fix the drainer
integration test so that we create separate codecs for the goroutines that the
test helper spins up to simulate client updates.
This changeset also refactors the drainer integration test to bring it up to
current idioms and library usages, make assertions more clear, and reduce
duplication.
* api: enable support for setting original source alongside job
This PR adds support for setting job source material along with
the registration of a job.
This includes a new HTTP endpoint and a new RPC endpoint for
making queries for the original source of a job. The
HTTP endpoint is /v1/job/<id>/submission?version=<version> and
the RPC method is Job.GetJobSubmission.
The job source (if submitted, and doing so is always optional), is
stored in the job_submission memdb table, separately from the
actual job. This way we do not incur overhead of reading the large
string field throughout normal job operations.
The server config now includes job_max_source_size for configuring
the maximum size the job source may be, before the server simply
drops the source material. This should help prevent Bad Things from
happening when huge jobs are submitted. If the value is set to 0,
all job source material will be dropped.
* api: avoid writing var content to disk for parsing
* api: move submission validation into RPC layer
* api: return an error if updating a job submission without namespace or job id
* api: be exact about the job index we associate a submission with (modify)
* api: reword api docs scheduling
* api: prune all but the last 6 job submissions
* api: protect against nil job submission in job validation
* api: set max job source size in test server
* api: fixups from pr
The `ephemeral_disk` block's `migrate` field allows for best-effort migration of
the ephemeral disk data to new nodes. The documentation says the `migrate` field
is only respected if `sticky=true`, but in fact if client ACLs are not set the
data is migrated even if `sticky=false`.
The existing behavior when client ACLs are disabled has existed since the early
implementation, so "fixing" that case now would silently break backwards
compatibility. Additionally, having `migrate` not imply `sticky` seems
nonsensical: it suggests that if we place on a new node we migrate the data but
if we place on the same node, we throw the data away!
Update so that `migrate=true` implies `sticky=true` as follows:
* The failure mode when client ACLs are enabled comes from the server not passing
along a migration token. Update the server so that the server provides a
migration token whenever `migrate=true` and not just when `sticky=true` too.
* Update the scheduler so that `migrate` implies `sticky`.
* Update the client so that we check for `migrate || sticky` where appropriate.
* Refactor the E2E tests to move them off the old framework and make the intention
of the test more clear.
Requests without an ACL token that pass thru the client's HTTP API are treated
as though they come from the client itself. This allows bypass of ACLs on RPC
requests where ACL permissions are checked (like `Job.Register`). Invalid tokens
are correctly rejected.
Fix the bypass by only setting a client ID on the identity if we have a valid node secret.
Note that this changeset will break rate metrics for RPCs sent by clients
without a client secret such as `Node.GetClientAllocs`; these requests will be
recorded as anonymous.
Future work should:
* Ensure the node secret is sent with all client-driven RPCs except
`Node.Register` which is TOFU.
* Create a new `acl.ACL` object from client requests so that we
can enforce ACLs for all endpoints in a uniform way that's less error-prone.~
* Multiple instances of a periodic job are run simultaneously, when prohibit_overlap is true
Fixes#11052
When restoring periodic dispatcher, all periodic jobs are forced without checking for previous childre.
* Multiple instances of a periodic job are run simultaneously, when prohibit_overlap is true
Fixes#11052
When restoring periodic dispatcher, all periodic jobs are forced without checking for previous children.
* style: refactor force run function
* fix: remove defer and inline unlock for speed optimization
* Update nomad/leader.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* style: refactor tests to use must
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* Update nomad/leader_test.go
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* fix: move back from defer to calling unlock before returning.
createEval cant be called with the lock on
* style: refactor test to use must
* added new entry to changelog and update comments
---------
Co-authored-by: James Rasell <jrasell@hashicorp.com>
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
This changeset refactors the tests of the draining node watcher so that we don't
mock the node watcher's `Remove` and `Update` methods for its own tests. Instead
we'll mock the node watcher's dependencies (the job watcher and deadline
notifier) and now unit tests can cover the real code. This allows us to remove a
bunch of TODOs in `watch_nodes.go` around testing and clarify some important
behaviors:
* Nodes that are down or disconnected will still be watched until the scheduler
decides what to do with their allocations. This will drive the job watcher but
not the node watcher, and that lets the node watcher gracefully handle cases
where a heartbeat fails but the node heartbeats again before its allocs can be
evicted.
* Stop watching nodes that have been deleted. The blocking query for nodes set
the maximum index to the highest index of a node it found, rather than the
index of the nodes table. This misses updates to the index from deleting
nodes. This was done as an performance optimization to avoid excessive
unblocking, but because the query is over all nodes anyways there's no
optimization to be had here. Remove the optimization so we can detect deleted
nodes without having to wait for an update to an unrelated node.
Fixes#16517
Given a 3 Server cluster with at least 1 Client connected to Follower 1:
If a NodeMeta.{Apply,Read} for the Client request is received by
Follower 1 with `AllowStale = false` the Follower will forward the
request to the Leader.
The Leader, not being connected to the target Client, will forward the
RPC to Follower 1.
Follower 1, seeing AllowStale=false, will forward the request to the
Leader.
The Leader, not being connected to... well hoppefully you get the
picture: an infinite loop occurs.
The job evaluate endpoint creates a new evaluation for the job which is
a write operation. This change modifies the necessary capability from
`read-job` to `submit-job` to better reflect this.
The tests for the system allocs reconciling code path (`diffSystemAllocs`)
include many impossible test environments, such as passing allocs for the wrong
node into the function. This makes the test assertions nonsensible for use in
walking yourself through the correct behavior.
I've pulled this changeset out of PR #16097 so that we can merge these
improvements and revisit the right approach to fix the problem in #16097 with
less urgency now that the PFNR bug fix has been merged. This changeset breaks up
a couple of tests, expands test coverage, and makes test assertions more
clear. It also corrects one bit of production code that behaves fine in
production because of canonicalization, but forces us to remember to set values
in tests to compensate.
In preperation for some refactoring to tasksUpdated, add a benchmark to the
old code so it's easy to compare with the changes, making sure nothing goes
off the rails for performance.
ACL policies can be associated with a job so that the job's Workload Identity
can have expanded access to other policy objects, including other
variables. Policies set on the variables the job automatically has access to
were ignored, but this includes policies with `deny` capabilities.
Additionally, when resolving claims for a workload identity without any attached
policies, the `ResolveClaims` method returned a `nil` ACL object, which is
treated similarly to a management token. While this was safe in Nomad 1.4.x,
when the workload identity token was exposed to the task via the `identity`
block, this allows a user with `submit-job` capabilities to escalate their
privileges.
We originally implemented automatic workload access to Variables as a separate
code path in the Variables RPC endpoint so that we don't have to generate
on-the-fly policies that blow up the ACL policy cache. This is fairly brittle
but also the behavior around wildcard paths in policies different from the rest
of our ACL polices, which is hard to reason about.
Add an `ACLClaim` parameter to the `AllowVariableOperation` method so that we
can push all this logic into the `acl` package and the behavior can be
consistent. This will allow a `deny` policy to override automatic access (and
probably speed up checks of non-automatic variable access).
When the scheduler tries to find a placement for a new allocation, it iterates
over a subset of nodes. For each node, we populate a `NetworkIndex` bitmap with
the ports of all existing allocations and any other allocations already proposed
as part of this same evaluation via its `SetAllocs` method. Then we make an
"ask" of the `NetworkIndex` in `AssignPorts` for any ports we need and receive
an "offer" in return. The offer will include both static ports and any dynamic
port assignments.
The `AssignPorts` method was written to support group networks, and it shares
code that selects dynamic ports with the original `AssignTaskNetwork`
code. `AssignTaskNetwork` can request multiple ports from the bitmap at a
time. But `AssignPorts` requests them one at a time and does not account for
possible collisions, and doesn't return an error in that case.
What happens next varies:
1. If the scheduler doesn't place the allocation on that node, the port
conflict is thrown away and there's no problem.
2. If the node is picked and this is the only allocation (or last allocation),
the plan applier will reject the plan when it calls `SetAllocs`, as we'd expect.
3. If the node is picked and there are additional allocations in the same eval
that iterate over the same node, their call to `SetAllocs` will detect the
impossible state and the node will be rejected. This can have the puzzling
behavior where a second task group for the job without any networking at all
can hit a port collision error!
It looks like this bug has existed since we implemented group networks, but
there are several factors that add up to making the issue rare for many users
yet frustratingly frequent for others:
* You're more likely to hit this bug the more tightly packed your range for
dynamic ports is. With 12000 ports in the range by default, many clusters can
avoid this for a long time.
* You're more likely to hit case (3) for jobs with lots of allocations or if a
scheduler has to iterate over a large number of nodes, such as with system jobs,
jobs with `spread` blocks, or (sometimes) jobs using `unique` constraints.
For unlucky combinations of these factors, it's possible that case (3) happens
repeatedly, preventing scheduling of a given job until a client state
change (ex. restarting the agent so all its allocations are rescheduled
elsewhere) re-opens the range of dynamic ports available.
This changeset:
* Fixes the bug by accounting for collisions in dynamic port selection in
`AssignPorts`.
* Adds test coverage for `AssignPorts`, expands coverage of this case for the
deprecated `AssignTaskNetwork`, and tightens the dynamic port range in a
scheduler test for spread scheduling to more easily detect this kind of problem
in the future.
* Adds a `String()` method to `Bitmap` so that any future "screaming" log lines
have a human-readable list of used ports.
Wildcard datacenters introduced a bug where a job with any wildcard datacenters
will always be treated as a destructive update when we check whether a
datacenter has been removed from the jobspec.
Includes updating the helper so that callers don't have to loop over the job's
datacenters.
In Nomad 0.12.1 we introduced atomic job registration/deregistration, where the
new eval was written in the same raft entry. Backwards-compatibility checks were
supposed to have been removed in Nomad 1.1.0, but we missed that. This is long
safe to remove.
When native service discovery was added, we used the node secret as the auth
token. Once Workload Identity was added in Nomad 1.4.x we needed to use the
claim token for `template` blocks, and so we allowed valid claims to bypass the
ACL policy check to preserve the existing behavior. (Invalid claims are still
rejected, so this didn't widen any security boundary.)
In reworking authentication for 1.5.0, we unintentionally removed this
bypass. For WIs without a policy attached to their job, everything works as
expected because the resulting `acl.ACL` is nil. But once a policy is attached
to the job the `acl.ACL` is no longer nil and this causes permissions errors.
Fix the regression by adding back the bypass for valid claims. In future work,
we should strongly consider getting turning the implicit policies into real
`ACLPolicy` objects (even if not stored in state) so that we don't have these
kind of brittle exceptions to the auth code.
The signature of the `raftApply` function requires that the caller unwrap the
first returned value (the response from `FSM.Apply`) to see if it's an
error. This puts the burden on the caller to remember to check two different
places for errors, and we've done so inconsistently.
Update `raftApply` to do the unwrapping for us and return any `FSM.Apply` error
as the error value. Similar work was done in Consul in
https://github.com/hashicorp/consul/pull/9991. This eliminates some boilerplate
and surfaces a few minor bugs in the process:
* job deregistrations of already-GC'd jobs were still emitting evals
* reconcile job summaries does not return scheduler errors
* node updates did not report errors associated with inconsistent service
discovery or CSI plugin states
Note that although _most_ of the `FSM.Apply` functions return only errors (which
makes it tempting to remove the first return value entirely), there are few that
return `bool` for some reason and Variables relies on the response value for
proper CAS checking.
This change resolves policies for workload identities when calling Client RPCs. Previously only ACL tokens could be used for Client RPCs.
Since the same cache is used for both bearer tokens (ACL and Workload ID), the token cache size was doubled.
---------
Co-authored-by: James Rasell <jrasell@users.noreply.github.com>
* build: add BuildDate to version info
will be used in enterprise to compare to license expiration time
* cli: multi-line version output, add BuildDate
before:
$ nomad version
Nomad v1.4.3 (coolfakecommithashomgoshsuchacoolonewoww)
after:
$ nomad version
Nomad v1.5.0-dev
BuildDate 2023-02-17T19:29:26Z
Revision coolfakecommithashomgoshsuchacoolonewoww
compare consul:
$ consul version
Consul v1.14.4
Revision dae670fe
Build Date 2023-01-26T15:47:10Z
Protocol 2 spoken by default, blah blah blah...
and vault:
$ vault version
Vault v1.12.3 (209b3dd99fe8ca320340d08c70cff5f620261f9b), built 2023-02-02T09:07:27Z
* docs: update version command output
The `CSIVolume` struct has references to allocations that are "denormalized"; we
don't store them on the `CSIVolume` struct but hydrate them on read. Tests
detecting potential state store corruptions found two locations where we're not
copying the volume before denormalizing:
* When garbage collecting CSI volume claims.
* When checking if it's safe to force-deregister the volume.
There are no known user-visible problems associated with these bugs but both
have the potential of mutating volume claims outside of a FSM transaction. This
changeset also cleans up state mutations in some CSI tests so as to avoid having
working tests cover up potential future bugs.
The RPC TLS enforcement test was frequently failing with broken connections. The
most likely cause was that the tests started to run before the server had
started its RPC server. Wait until it self-elects to ensure that the RPC server
is up. This seems to have corrected the error; I ran this 3 times without a
failure (even accounting for `gotestsum` retries).
Also, fix a minor test bug that didn't impact the test but showed an incorrect
usage for `Status.Ping.`
Tim Gross