When a Nomad client that is running an allocation with
`max_client_disconnect` set misses a heartbeat the Nomad server will
update its status to `disconnected`.
Upon reconnecting, the client will make three main RPC calls:
- `Node.UpdateStatus` is used to set the client status to `ready`.
- `Node.UpdateAlloc` is used to update the client-side information about
allocations, such as their `ClientStatus`, task states etc.
- `Node.Register` is used to upsert the entire node information,
including its status.
These calls are made concurrently and are also running in parallel with
the scheduler. Depending on the order they run the scheduler may end up
with incomplete data when reconciling allocations.
For example, a client disconnects and its replacement allocation cannot
be placed anywhere else, so there's a pending eval waiting for
resources.
When this client comes back the order of events may be:
1. Client calls `Node.UpdateStatus` and is now `ready`.
2. Scheduler reconciles allocations and places the replacement alloc to
the client. The client is now assigned two allocations: the original
alloc that is still `unknown` and the replacement that is `pending`.
3. Client calls `Node.UpdateAlloc` and updates the original alloc to
`running`.
4. Scheduler notices too many allocs and stops the replacement.
This creates unnecessary placements or, in a different order of events,
may leave the job without any allocations running until the whole state
is updated and reconciled.
To avoid problems like this clients must update _all_ of its relevant
information before they can be considered `ready` and available for
scheduling.
To achieve this goal the RPC endpoints mentioned above have been
modified to enforce strict steps for nodes reconnecting:
- `Node.Register` does not set the client status anymore.
- `Node.UpdateStatus` sets the reconnecting client to the `initializing`
status until it successfully calls `Node.UpdateAlloc`.
These changes are done server-side to avoid the need of additional
coordination between clients and servers. Clients are kept oblivious of
these changes and will keep making these calls as they normally would.
The verification of whether allocations have been updates is done by
storing and comparing the Raft index of the last time the client missed
a heartbeat and the last time it updated its allocations.
Implement a metric for RPC requests with labels on the identity, so that
administrators can monitor the source of requests within the cluster. This
changeset demonstrates the change with the new `ACL.WhoAmI` RPC, and we'll wire
up the remaining RPCs once we've threaded the new pre-forwarding authentication
through the all.
Note that metrics are measured after we forward but before we return any
authentication error. This ensures that we only emit metrics on the server that
actually serves the request. We'll perform rate limiting at the same place.
Includes telemetry configuration to omit identity labels.
* consul: correctly understand missing consul checks as unhealthy
This PR fixes a bug where Nomad assumed any registered Checks would exist
in the service registration coming back from Consul. In some cases, the
Consul may be slow in processing the check registration, and the response
object would not contain checks. Nomad would then scan the empty response
looking for Checks with failing health status, finding none, and then
marking a task/alloc as healthy.
In reality, we must always use Nomad's view of what checks should exist as
the source of truth, and compare that with the response Consul gives us,
making sure they match, before scanning the Consul response for failing
check statuses.
Fixes#15536
* consul: minor CR refactor using maps not sets
* consul: observe transition from healthy to unhealthy checks
* consul: spell healthy correctly
This changeset covers a sidebar discussion that @schmichael and I had around the
design for pre-forwarding auth. This includes some changes extracted out of
#15513 to make it easier to review both and leave a clean history.
* Remove fast path for NodeID. Previously-connected clients will have a NodeID
set on the context, and because this is a large portion of the RPCs sent we
fast-pathed it at the top of the `Authenticate` method. But the context is
shared for all yamux streams over the same yamux session (and TCP
connection). This lets an authenticated HTTP request to a client use the
NodeID for authentication, which is a privilege escalation. Remove the fast
path and annotate it so that we don't break it again.
* Add context to decisions around AuthenticatedIdentity. The `Authenticate`
method taken on its own looks like it wants to return an `acl.ACL` that folds
over all the various identity types (creating an ephemeral ACL on the fly if
neccessary). But keeping these fields idependent allows RPC handlers to
differentiate between internal and external origins so we most likely want to
avoid this. Leave some docstrings as a warning as to why this is built the way
it is.
* Mutate the request rather than returning. When reviewing #15513 we decided
that forcing the request handler to call `SetIdentity` was repetitive and
error prone. Instead, the `Authenticate` method mutates the request by setting
its `AuthenticatedIdentity`.
This change adds a new table that will store ACL binding rule
objects. The two indexes allow fast lookups by their ID, or by
which auth method they are linked to. Snapshot persist and
restore functionality ensures this table can be saved and
restored from snapshots.
In order to write and delete the object to state, new Raft messages
have been added.
All RPC request and response structs, along with object functions
such as diff and canonicalize have been included within this work
as it is nicely separated from the other areas of work.
Upcoming work to instrument the rate of RPC requests by consumer (and eventually
rate limit) require that we authenticate a RPC request before forwarding. Add a
new top-level `Authenticate` method to the server and have it return an
`AuthenticatedIdentity` struct. RPC handlers will use the relevant fields of
this identity for performing authorization.
This changeset includes:
* The main implementation of `Authenticate`
* Provide a new RPC `ACL.WhoAmI` for debugging authentication. This endpoint
returns the same `AuthenticatedIdentity` that will be used by RPC handlers. At
some point we might want to give this an equivalent HTTP endpoint but I didn't
want to add that to our public API until some of the other Workload Identity
work is solidified, especially if we don't need it yet.
* A full coverage test of the `Authenticate` method. This sets up two server
nodes with mTLS and ACLs, some tokens, and some allocations with workload
identities.
* Wire up an example of using `Authenticate` in the `Namespace.Upsert` RPC and
see how authorization happens after forwarding.
* A new semgrep rule for `Authenticate`, which we'll need to update once we're
ready to wire up more RPC endpoints with authorization steps.
* scheduler: create placements for non-register MRD
For multiregion jobs, the scheduler does not create placements on
registration because the deployment must wait for the other regions.
Once of these regions will then trigger the deployment to run.
Currently, this is done in the scheduler by considering any eval for a
multiregion job as "paused" since it's expected that another region will
eventually unpause it.
This becomes a problem where evals not triggered by a job registration
happen, such as on a node update. These types of regional changes do not
have other regions waiting to progress the deployment, and so they were
never resulting in placements.
The fix is to create a deployment at job registration time. This
additional piece of state allows the scheduler to differentiate between
a multiregion change, where there are other regions engaged in the
deployment so no placements are required, from a regional change, where
the scheduler does need to create placements.
This deployment starts in the new "initializing" status to signal to the
scheduler that it needs to compute the initial deployment state. The
multiregion deployment will wait until this deployment state is
persisted and its starts is set to "pending". Without this state
transition it's possible to hit a race condition where the plan applier
and the deployment watcher may step of each other and overwrite their
changes.
* changelog: add entry for #15325
During unusual outage recovery scenarios on large clusters, a backlog of
millions of evaluations can appear. In these cases, the `eval delete` command can
put excessive load on the cluster by listing large sets of evals to extract the
IDs and then sending larges batches of IDs. Although the command's batch size
was carefully tuned, we still need to be JSON deserialize, re-serialize to
MessagePack, send the log entries through raft, and get the FSM applied.
To improve performance of this recovery case, move the batching process into the
RPC handler and the state store. The design here is a little weird, so let's
look a the failed options first:
* A naive solution here would be to just send the filter as the raft request and
let the FSM apply delete the whole set in a single operation. Benchmarking with
1M evals on a 3 node cluster demonstrated this can block the FSM apply for
several minutes, which puts the cluster at risk if there's a leadership
failover (the barrier write can't be made while this apply is in-flight).
* A less naive but still bad solution would be to have the RPC handler filter
and paginate, and then hand a list of IDs to the existing raft log
entry. Benchmarks showed this blocked the FSM apply for 20-30s at a time and
took roughly an hour to complete.
Instead, we're filtering and paginating in the RPC handler to find a page token,
and then passing both the filter and page token in the raft log. The FSM apply
recreates the paginator using the filter and page token to get roughly the same
page of evaluations, which it then deletes. The pagination process is fairly
cheap (only abut 5% of the total FSM apply time), so counter-intuitively this
rework ends up being much faster. A benchmark of 1M evaluations showed this
blocked the FSM apply for 20-30ms at a time (typical for normal operations) and
completes in less than 4 minutes.
Note that, as with the existing design, this delete is not consistent: a new
evaluation inserted "behind" the cursor of the pagination will fail to be
deleted.
This PR implements ACLAuthMethod type, acl_auth_methods table schema and crud state store methods. It also updates nomadSnapshot.Persist and nomadSnapshot.Restore methods in order for them to work with the new table, and adds two new Raft messages: ACLAuthMethodsUpsertRequestType and ACLAuthMethodsDeleteRequestType
This PR is part of the SSO work captured under ☂️ ticket #13120.
Add a new `Eval.Count` RPC and associated HTTP API endpoints. This API is
designed to support interactive use in the `nomad eval delete` command to get a
count of evals expected to be deleted before doing so.
The state store operations to do this sort of thing are somewhat expensive, but
it's cheaper than serializing a big list of evals to JSON. Note that although it
seems like this could be done as an extra parameter and response field on
`Eval.List`, having it as its own endpoint avoids having to change the response
body shape and lets us avoid handling the legacy filter params supported by
`Eval.List`.
* 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
* Adds meta to job list stub and displays a pack logo on the jobs index
* Changelog
* Modifying struct for optional meta param
* Explicitly ask for meta anytime I look up a job from index or job page
* Test case for the endpoint
* adding meta field to API struct and ommitting from response if empty
* passthru method added to api/jobs.list
* Meta param listed in docs for jobs list
* Update api/jobs.go
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Co-authored-by: Tim Gross <tgross@hashicorp.com>
When replication of a single key fails, the replication loop breaks early and
therefore keys that fall later in the sorting order will never get
replicated. This is particularly a problem for clusters impacted by the bug that
caused #14981 and that were later upgraded; the keys that were never replicated
can now never be replicated, and so we need to handle them safely.
Included in the replication fix:
* Refactor the replication loop so that each key replicated in a function call
that returns an error, to make the workflow more clear and reduce nesting. Log
the error and continue.
* Improve stability of keyring replication tests. We no longer block leadership
on initializing the keyring, so there's a race condition in the keyring tests
where we can test for the existence of the root key before the keyring has
been initialize. Change this to an "eventually" test.
But these fixes aren't enough to fix#14981 because they'll end up seeing an
error once a second complaining about the missing key, so we also need to fix
keyring GC so the keys can be removed from the state store. Now we'll store the
key ID used to sign a workload identity in the Allocation, and we'll index the
Allocation table on that so we can track whether any live Allocation was signed
with a particular key ID.
ACL tokens are granted permissions either by direct policy links
or via ACL role links. Callers should therefore be able to read
policies directly assigned to the caller token or indirectly by
ACL role links.
This PR removes the assertion around when the 'task' field of
a check may be set. Starting in Nomad 1.4 we automatically set
the task field on all checks in support of the NSD checks feature.
This is causing validation problems elsewhere, e.g. when a group
service using the Consul provider sets 'task' it will fail
validation that worked previously.
The assertion of leaving 'task' unset was only about making sure
job submitters weren't expecting some behavior, but in practice
is causing bugs now that we need the task field for more than it
was originally added for.
We can simply update the docs, noting when the task field set by
job submitters actually has value.
* scheduler: Fix bug where the scheduler would treat multiregion jobs as paused for job types that don't use deployments
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Co-authored-by: Tim Gross <tgross@hashicorp.com>
* 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
* 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
The current implementation for the task coordinator unblocks tasks by
performing destructive operations over its internal state (like closing
channels and deleting maps from keys).
This presents a problem in situations where we would like to revert the
state of a task, such as when restarting an allocation with tasks that
have already exited.
With this new implementation the task coordinator behaves more like a
finite state machine where task may be blocked/unblocked multiple times
by performing a state transition.
This initial part of the work only refactors the task coordinator and
is functionally equivalent to the previous implementation. Future work
will build upon this to provide bug fixes and enhancements.
The original design for workload identities and ACLs allows for operators to
extend the automatic capabilities of a workload by using a specially-named
policy. This has shown to be potentially unsafe because of naming collisions, so
instead we'll allow operators to explicitly attach a policy to a workload
identity.
This changeset adds workload identity fields to ACL policy objects and threads
that all the way down to the command line. It also a new secondary index to the
ACL policy table on namespace and job so that claim resolution can efficiently
query for related policies.
When a Nomad agent starts and loads jobs that already existed in the
cluster, the default template uid and gid was being set to 0, since this
is the zero value for int. This caused these jobs to fail in
environments where it was not possible to use 0, such as in Windows
clients.
In order to differentiate between an explicit 0 and a template where
these properties were not set we need to use a pointer.
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!"
ACL tokens can now utilize ACL roles in order to provide API
authorization. Each ACL token can be created and linked to an
array of policies as well as an array of ACL role links. The link
can be provided via the role name or ID, but internally, is always
resolved to the ID as this is immutable whereas the name can be
changed by operators.
When resolving an ACL token, the policies linked from an ACL role
are unpacked and combined with the policy array to form the
complete auth set for the token.
The ACL token creation endpoint handles deduplicating ACL role
links as well as ensuring they exist within state.
When reading a token, Nomad will also ensure the ACL role link is
current. This handles ACL roles being deleted from under a token
from a UX standpoint.
Move conflict resolution implementation into the state store with a new Apply RPC.
This also makes the RPC for secure variables much more similar to Consul's KV,
which will help us support soft deletes in a post-1.4.0 version of Nomad.
Reimplement quotas in the state store functions.
Co-authored-by: Charlie Voiselle <464492+angrycub@users.noreply.github.com>
This commit includes the new state schema for ACL roles along with
state interaction functions for CRUD actions.
The change also includes snapshot persist and restore
functionality and the addition of FSM messages for Raft updates
which will come via RPC endpoints.
Workload identities grant implicit access to policies, and operators
will not want to craft separate policies for each invocation of a
periodic or dispatch job. Use the parent job's ID as the JobID claim.
Piotr Kazmierczak