When a disconnect client reconnects the `allocReconciler` must find the
allocations that were created to replace the original disconnected
allocations.
This process was being done in only a subset of non-terminal untainted
allocations, meaning that, if the replacement allocations were not in
this state the reconciler didn't stop them, leaving the job in an
inconsistent state.
This inconsistency is only solved in a future job evaluation, but at
that point the allocation is considered reconnected and so the specific
reconnection logic was not applied, leading to unexpected outcomes.
This commit fixes the problem by running reconnecting allocation
reconciliation logic earlier into the process, leaving the rest of the
reconciler oblivious of reconnecting allocations.
It also uses the full set of allocations to search for replacements,
stopping them even if they are not in the `untainted` set.
The system `SystemScheduler` is not affected by this bug because
disconnected clients don't trigger replacements: every eligible client
is already running an allocation.
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.
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.
* main: remove deprecated uses of rand.Seed
go1.20 deprecates rand.Seed, and seeds the rand package
automatically. Remove cases where we seed the random package,
and cleanup the one case where we intentionally create a
known random source.
* cl: update cl
* mod: update go mod
Many of the functions in the `utils.go` file are specific to a particular
scheduler, and very few of them have guards (or even names) that help avoid
misuse with features specific to a given scheduler type. Move these
functions (and their tests) into files specific to their scheduler type without
any functionality changes to make it clear which bits go with what.
Service jobs should have unique allocation Names, derived from the
Job.ID. System jobs do not have unique allocation Names because the index is
intended to indicated the instance out of a desired count size. Because system
jobs do not have an explicit count but the results are based on the targeted
nodes, the index is less informative and this was intentionally omitted from the
original design.
Update docs to make it clear that NOMAD_ALLOC_INDEX is always zero for
system/sysbatch jobs
Validate that `volume.per_alloc` is incompatible with system/sysbatch jobs.
System and sysbatch jobs always have a `NOMAD_ALLOC_INDEX` of 0. So
interpolation via `per_alloc` will not work as soon as there's more than one
allocation placed. Validate against this on job submission.
Add `identity` jobspec block to expose workload identity tokens to tasks.
---------
Co-authored-by: Anders <mail@anars.dk>
Co-authored-by: Tim Gross <tgross@hashicorp.com>
Co-authored-by: Michael Schurter <mschurter@hashicorp.com>
Disallowing per_alloc for host volumes in some cases makes life of a nomad user much harder.
When we rely on the NOMAD_ALLOC_INDEX for any configuration that needs to be re-used across
restarts we need to make sure allocation placement is consistent. With CSI volumes we can
use the `per_alloc` feature but for some reason this is explicitly disabled for host volumes.
Ensure host volumes understand the concept of per_alloc
Devices are fingerprinted as groups of similar devices. This prevented
specifying specific device by their ID in constraint and affinity rules.
This commit introduces the `${device.ids}` attribute that returns a
comma separated list of IDs that are part of the device group. Users can
then use the set operators to write rules.
* 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
When the scheduler checks feasibility of each node, it creates a "stack" which
carries attributes of the job and task group it needs to check feasibility
for. The `system` and `sysbatch` scheduler use a different stack than `service`
and `batch` jobs. This stack was missing the call to set the job ID and
namespace for the CSI check. This prevents CSI volumes from being scheduled for
system jobs whenever the volume is in a non-default namespace.
Set the job ID and namespace to match the generic scheduler.
* 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
* One-time tokens are not replicated between regions, so we don't want to enforce
that the version check across all of serf, just members in the same region.
* Scheduler: Disconnected clients handling is specific to a single region, so we
don't want to enforce that the version check across all of serf, just members in
the same region.
* Variables: enforce version check in Apply RPC
* Cleans up a bunch of legacy checks.
This changeset is specific to 1.4.x and the changes for previous versions of
Nomad will be manually backported in a separate PR.
* cleanup: fixup linter warnings in schedular/feasible.go
* core: numeric operands comparisons in constraints
This PR changes constraint comparisons to be numeric rather than
lexical if both operands are integers or floats.
Inspiration #4856Closes#4729Closes#14719
* fix: always parse as int64
* 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
* scheduler: stopped-yet-running allocs are still running
* scheduler: test new stopped-but-running logic
* test: assert nonoverlapping alloc behavior
Also add a simpler Wait test helper to improve line numbers and save few
lines of code.
* docs: tried my best to describe #10446
it's not concise... feedback welcome
* scheduler: fix test that allowed overlapping allocs
* devices: only free devices when ClientStatus is terminal
* test: output nicer failure message if err==nil
Co-authored-by: Mahmood Ali <mahmood@hashicorp.com>
Co-authored-by: Michael Schurter <mschurter@hashicorp.com>
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.
Fixes#13505
This fixes#13505 by treating reserved_ports like we treat a lot of jobspec settings: merging settings from more global stanzas (client.reserved.reserved_ports) "down" into more specific stanzas (client.host_networks[].reserved_ports).
As discussed in #13505 there are other options, and since it's totally broken right now we have some flexibility:
Treat overlapping reserved_ports on addresses as invalid and refuse to start agents. However, I'm not sure there's a cohesive model we want to publish right now since so much 0.9-0.12 compat code still exists! We would have to explain to folks that if their -network-interface and host_network addresses overlapped, they could only specify reserved_ports in one place or the other?! It gets ugly.
Use the global client.reserved.reserved_ports value as the default and treat host_network[].reserverd_ports as overrides. My first suggestion in the issue, but @groggemans made me realize the addresses on the agent's interface (as configured by -network-interface) may overlap with host_networks, so you'd need to remove the global reserved_ports from addresses shared with a shared network?! This seemed really confusing and subtle for users to me.
So I think "merging down" creates the most expressive yet understandable approach. I've played around with it a bit, and it doesn't seem too surprising. The only frustrating part is how difficult it is to observe the available addresses and ports on a node! However that's a job for another PR.
Stream snapshot to FSM when restoring from archive
The `RestoreFromArchive` helper decompresses the snapshot archive to a
temporary file before reading it into the FSM. For large snapshots
this performs a lot of disk IO. Stream decompress the snapshot as we
read it, without first writing to a temporary file.
Add bexpr filters to the `RestoreFromArchive` helper.
The operator can pass these as `-filter` arguments to `nomad operator
snapshot state` (and other commands in the future) to include only
desired data when reading the snapshot.
As a performance optimization in the scheduler, feasibility checks
that apply to an entire class are only checked once for all nodes of
that class. Other feasibility checks are "available" checks because
they rely on more ephemeral characteristics and don't contribute to
the hash for the node class. This currently includes only CSI.
We have a separate fast path for "available" checks when the node has
already been marked eligible on the basis of class. This fast path has
a bug where it returns early rather than continuing the loop. This
causes the entire task group to be rejected.
Fix the bug by not returning early in the fast path and instead jump
to the top of the loop like all the other code paths in this method.
Includes a new test exercising topology at whole-scheduler level and a
fix for an existing test that should've caught this previously.
In the reconciler's filtering for tainted nodes, we use whether the
server supports disconnected clients as a gate to a bunch of our
logic, but this doesn't account for cases where the job doesn't have
`max_client_disconnect`. The only real consequence of this appears to
be that allocs on disconnected nodes are marked "complete" instead of
"lost".
* planner: expose ServerMeetsMinimumVersion via Planner interface
* filterByTainted: add flag indicating disconnect support
* allocReconciler: accept and pass disconnect support flag
* tests: update dependent tests
