* client/heartbeatstop: reversed time condition for startup grace
* scheduler/generic_sched: use `delayInstead` to avoid a loop
Without protecting the loop that creates followUpEvals, a delayed eval
is allowed to create an immediate subsequent delayed eval. For both
`stop_after_client_disconnect` and the `reschedule` block, a delayed
eval should always produce some immediate result (running or blocked)
and then only after the outcome of that eval produce a second delayed
eval.
* scheduler/reconcile: lostLater are different than delayedReschedules
Just slightly. `lostLater` allocs should be used to create batched
evaluations, but `handleDelayedReschedules` assumes that the
allocations are in the untainted set. When it creates the in-place
updates to those allocations at the end, it causes the allocation to
be treated as running over in the planner, which causes the initial
`stop_after_client_disconnect` evaluation to be retried by the worker.
* changes necessary to support oss licesning shims
revert nomad fmt changes
update test to work with enterprise changes
update tests to work with new ent enforcements
make check
update cas test to use scheduler algorithm
back out preemption changes
add comments
* remove unused method
This fixes few cases where driver eventor goroutines are leaked during
normal operations, but especially so in tests.
This change makes few modifications:
First, it switches drivers to use `Context`s to manage shutdown events.
Previously, it relied on callers invoking `.Shutdown()` function that is
specific to internal drivers only and require casting. Using `Contexts`
provide a consistent idiomatic way to manage lifecycle for both internal
and external drivers.
Also, I discovered few places where we don't clean up a temporary driver
instance in the plugin catalog code, where we dispense a driver to
inspect and validate the schema config without properly cleaning it up.
When an allocation runs for a task driver that can't support volume mounts,
the mounting will fail in a way that can be hard to understand. With host
volumes this usually means failing silently, whereas with CSI the operator
gets inscrutable internals exposed in the `nomad alloc status`.
This changeset adds a MountConfig field to the task driver Capabilities
response. We validate this when the `csi_hook` or `volume_hook` fires and
return a user-friendly error.
Note that we don't currently have a way to get driver capabilities up to the
server, except through attributes. Validating this when the user initially
submits the jobspec would be even better than what we're doing here (and could
be useful for all our other capabilities), but that's out of scope for this
changeset.
Also note that the MountConfig enum starts with "supports all" in order to
support community plugins in a backwards compatible way, rather than cutting
them off from volume mounting unexpectedly.
The `stats_hook` writes an Error log every time an allocation becomes
terminal. This is a normal condition, not an error. A real error
condition like a failure to collect the stats is logged later. It just
creates log noise, and this is a particularly bad operator experience
for heavy batch workloads.
The plugin supervisor lazily connects to plugins, but this means we
only get "Unavailable" back from the gRPC call in cases where the
plugin can never be reached (for example, if the Nomad client has the
wrong permissions for the socket).
This changeset improves the operator experience by switching to a
blocking `DialWithContext`. It eagerly connects so that we can
validate the connection is real and get a "failed to open" error in
case where Nomad can't establish the initial connection.
The MVP for CSI in the 0.11.0 release of Nomad did not include support
for opaque volume parameters or volume context. This changeset adds
support for both.
This also moves args for ControllerValidateCapabilities into a struct.
The CSI plugin `ControllerValidateCapabilities` struct that we turn
into a CSI RPC is accumulating arguments, so moving it into a request
struct will reduce the churn of this internal API, make the plugin
code more readable, and make this method consistent with the other
plugin methods in that package.
The plugin supervisor lazily connects to plugins, but this means we
only get "Unavailable" back from the gRPC call in cases where the
plugin can never be reached (for example, if the Nomad client has the
wrong permissions for the socket).
This changeset improves the operator experience by switching to a
blocking `DialWithContext`. It eagerly connects so that we can
validate the connection is real and get a "failed to open" error in
case where Nomad can't establish the initial connection.
The CSI plugins RPCs require the use of the storage provider's volume
ID, rather than the user-defined volume ID. Although changing the RPCs
to use the field name `ExternalID` risks breaking backwards
compatibility, we can use the `ExternalID` name internally for the
client and only use `VolumeID` at the RPC boundaries.
* jobspec, api: add stop_after_client_disconnect
* nomad/state/state_store: error message typo
* structs: alloc methods to support stop_after_client_disconnect
1. a global AllocStates to track status changes with timestamps. We
need this to track the time at which the alloc became lost
originally.
2. ShouldClientStop() and WaitClientStop() to actually do the math
* scheduler/reconcile_util: delayByStopAfterClientDisconnect
* scheduler/reconcile: use delayByStopAfterClientDisconnect
* scheduler/util: updateNonTerminalAllocsToLost comments
This was setup to only update allocs to lost if the DesiredStatus had
already been set by the scheduler. It seems like the intention was to
update the status from any non-terminal state, and not all lost allocs
have been marked stop or evict by now
* scheduler/testing: AssertEvalStatus just use require
* scheduler/generic_sched: don't create a blocked eval if delayed
* scheduler/generic_sched_test: several scheduling cases
CSI plugins can require credentials for some publishing and
unpublishing workflow RPCs. Secrets are configured at the time of
volume registration, stored in the volume struct, and then passed
around as an opaque map by Nomad to the plugins.
When serializing structs with msgpack, only consider type tags of
`codec`.
Hashicorp/go-msgpack (based on ugorji/go) defaults to interpretting
`codec` tag if it's available, but falls to using `json` if `codec`
isn't present.
This behavior is surprising in cases where we want to serialize json
differently from msgpack, e.g. serializing `ConsulExposeConfig`.
This change deflakes TestTaskTemplateManager_BlockedEvents test, because
it is expecting a number of events without accounting for transitional
state.
The test TestTaskTemplateManager_BlockedEvents attempts to ensure that a
template rendering emits blocked events for missing template ksys.
It works by setting a template that requires keys 0,1,2,3,4 and then
eventually sets keys 0,1,2,3 and ensures that we get a final event indicating
that keys 3 and 4 are still missing.
The test waits to get a blocked event for the final state, but it can
fail if receives a blocked event for a transitional state (e.g. one
reporting 2,3,4,5 are missing).
This fixes the test by ensuring that it waits until the final message
before assertion.
Also, it clarifies the intent of the test with stricter assertions and
additional comments.
Makes it possible to run Linux Containers On Windows with Nomad alongside Windows Containers. Fingerprint prevents only to run Nomad in Windows 10 with Linux Containers
In order to minimize this change while keeping a simple version of the
behavior, we set `lastOk` to the current time less the intial server
connection timeout. If the client starts and never contacts the
server, it will stop all configured tasks after the initial server
connection grace period, on the assumption that we've been out of
touch longer than any configured `stop_after_client_disconnect`.
The more complex state behavior might be justified later, but we
should learn about failure modes first.
- track lastHeartbeat, the client local time of the last successful
heartbeat round trip
- track allocations with `stop_after_client_disconnect` configured
- trigger allocation destroy (which handles cleanup)
- restore heartbeat/killable allocs tracking when allocs are recovered from disk
- on client restart, stop those allocs after a grace period if the
servers are still partioned
During MVP development, we reduced the timeout for controller plugins
to avoid long hangs in GC workers. But now that this work has been
moved to the volume watcher, we can restore the original timeout which
is better suited for the characteristic timescales of some cloud
provider APIs and better matches the behavior of k8s.
Fixes#7681
The current behavior of the CPU fingerprinter in AWS is that it
reads the **current** speed from `/proc/cpuinfo` (`CPU MHz` field).
This is because the max CPU frequency is not available by reading
anything on the EC2 instance itself. Normally on Linux one would
look at e.g. `sys/devices/system/cpu/cpuN/cpufreq/cpuinfo_max_freq`
or perhaps parse the values from the `CPU max MHz` field in
`/proc/cpuinfo`, but those values are not available.
Furthermore, no metadata about the CPU is made available in the
EC2 metadata service.
https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-categories.html
Since `go-psutil` cannot determine the max CPU speed it defaults to
the current CPU speed, which could be basically any number between
0 and the true max. This is particularly bad on large, powerful
reserved instances which often idle at ~800 MHz while Nomad does
its fingerprinting (typically IO bound), which Nomad then uses as
the max, which results in severe loss of available resources.
Since the CPU specification is unavailable programmatically (at least
not without sudo) use a best-effort lookup table. This table was
generated by going through every instance type in AWS documentation
and copy-pasting the numbers.
https://aws.amazon.com/ec2/instance-types/
This approach obviously is not ideal as future instance types will
need to be added as they are introduced to AWS. However, using the
table should only be an improvement over the status quo since right
now Nomad miscalculates available CPU resources on all instance types.
Use v1.1.5 of go-msgpack/codec/codecgen, so go-msgpack codecgen matches
the library version.
We branched off earlier to pick up
f51b518921
, but apparently that's not needed as we could customize the package via
`-c` argument.
Adds a `CSIVolumeClaim` type to be tracked as current and past claims
on a volume. Allows for a client RPC failure during node or controller
detachment without having to keep the allocation around after the
first garbage collection eval.
This changeset lays groundwork for moving the actual detachment RPCs
into a volume watching loop outside the GC eval.
task shutdown_delay will currently only run if there are registered
services for the task. This implementation detail isn't explicity stated
anywhere and is defined outside of the service stanza.
This change moves shutdown_delay to be evaluated after prekill hooks are
run, outside of any task runner hooks.
just use time.sleep
The `Job.Deregister` call will block on the client CSI controller RPCs
while the alloc still exists on the Nomad client node. So we need to
make the volume claim reaping async from the `Job.Deregister`. This
allows `nomad job stop` to return immediately. In order to make this
work, this changeset changes the volume GC so that the GC jobs are on a
by-volume basis rather than a by-job basis; we won't have to query
the (possibly deleted) job at the time of volume GC. We smuggle the
volume ID and whether it's a purge into the GC eval ID the same way we
smuggled the job ID previously.
The CSI plugins uses the external volume ID for all operations, but
the Client CSI RPCs uses the Nomad volume ID (human-friendly) for the
mount paths. Pass the External ID as an arg in the RPC call so that
the unpublish workflows have it without calling back to the server to
find the external ID.
The controller CSI plugins need the CSI node ID (or in other words,
the storage provider's view of node ID like the EC2 instance ID), not
the Nomad node ID, to determine how to detach the external volume.
If a volume-claiming alloc stops and the CSI Node plugin that serves
that alloc's volumes is missing, there's no way for the allocrunner
hook to send the `NodeUnpublish` and `NodeUnstage` RPCs.
This changeset addresses this issue with a redesign of the client-side
for CSI. Rather than unmounting in the alloc runner hook, the alloc
runner hook will simply exit. When the server gets the
`Node.UpdateAlloc` for the terminal allocation that had a volume claim,
it creates a volume claim GC job. This job will made client RPCs to a
new node plugin RPC endpoint, and only once that succeeds, move on to
making the client RPCs to the controller plugin. If the node plugin is
unavailable, the GC job will fail and be requeued.
Fixes#6594#6711#6714#7567
e2e testing is still TBD in #6502
Before, we only passed the Nomad agent's configured Consul HTTP
address onto the `consul connect envoy ...` bootstrap command.
This meant any Consul setup with TLS enabled would not work with
Nomad's Connect integration.
This change now sets CLI args and Environment Variables for
configuring TLS options for communicating with Consul when doing
the envoy bootstrap, as described in
https://www.consul.io/docs/commands/connect/envoy.html#usage
Enable configuration of HTTP and gRPC endpoints which should be exposed by
the Connect sidecar proxy. This changeset is the first "non-magical" pass
that lays the groundwork for enabling Consul service checks for tasks
running in a network namespace because they are Connect-enabled. The changes
here provide for full configuration of the
connect {
sidecar_service {
proxy {
expose {
paths = [{
path = <exposed endpoint>
protocol = <http or grpc>
local_path_port = <local endpoint port>
listener_port = <inbound mesh port>
}, ... ]
}
}
}
stanza. Everything from `expose` and below is new, and partially implements
the precedent set by Consul:
https://www.consul.io/docs/connect/registration/service-registration.html#expose-paths-configuration-reference
Combined with a task-group level network port-mapping in the form:
port "exposeExample" { to = -1 }
it is now possible to "punch a hole" through the network namespace
to a specific HTTP or gRPC path, with the anticipated use case of creating
Consul checks on Connect enabled services.
A future PR may introduce more automagic behavior, where we can do things like
1) auto-fill the 'expose.path.local_path_port' with the default value of the
'service.port' value for task-group level connect-enabled services.
2) automatically generate a port-mapping
3) enable an 'expose.checks' flag which automatically creates exposed endpoints
for every compatible consul service check (http/grpc checks on connect
enabled services).
* nomad/structs/structs: new NodeEventSubsystemCSI
* client/client: pass triggerNodeEvent in the CSIConfig
* client/pluginmanager/csimanager/instance: add eventer to instanceManager
* client/pluginmanager/csimanager/manager: pass triggerNodeEvent
* client/pluginmanager/csimanager/volume: node event on [un]mount
* nomad/structs/structs: use storage, not CSI
* client/pluginmanager/csimanager/volume: use storage, not CSI
* client/pluginmanager/csimanager/volume_test: eventer
* client/pluginmanager/csimanager/volume: event on error
* client/pluginmanager/csimanager/volume_test: check event on error
* command/node_status: remove an extra space in event detail format
* client/pluginmanager/csimanager/volume: use snake_case for details
* client/pluginmanager/csimanager/volume_test: snake_case details
The CSI Specification defines various gRPC Errors and how they may be retried. After auditing all our CSI RPC calls in #6863, this changeset:
* adds retries and backoffs to the where they were needed but not implemented
* annotates those CSI RPCs that do not need retries so that we don't wonder whether it's been left off accidentally
* added a timeout and cancellation context to the `Probe` call, which didn't have one.
The test inserts an alloc in the server state, but expect the client to
start the alloc runner for it almost immediately.
Here, we add a retry loop to check that the client start all expected
alloc runners eventually.
Fix a regression where we accidentally started treating non-AWS
environments as AWS environments, resulting in bad networking settings.
Two factors some at play:
First, in [1], we accidentally switched the ultimate AWS test from
checking `ami-id` to `instance-id`. This means that nomad started
treating more environments as AWS; e.g. Hetzner implements `instance-id`
but not `ami-id`.
Second, some of these environments return empty values instead of
errors! Hetzner returns empty 200 response for `local-ipv4`, resulting
into bad networking configuration.
This change fix the situation by restoring the check to `ami-id` and
ensuring that we only set network configuration when the ip address is
not-empty. Also, be more defensive around response whitespace input.
[1] https://github.com/hashicorp/nomad/pull/6779
Add mount_options to both the volume definition on registration and to the volume block in the group where the volume is requested. If both are specified, the options provided in the request replace the options defined in the volume. They get passed to the NodePublishVolume, which causes the node plugin to actually mount the volume on the host.
Individual tasks just mount bind into the host mounted volume (unchanged behavior). An operator can mount the same volume with different options by specifying it twice in the group context.
closes#7007
* nomad/structs/volumes: add MountOptions to volume request
* jobspec/test-fixtures/basic.hcl: add mount_options to volume block
* jobspec/parse_test: add expected MountOptions
* api/tasks: add mount_options
* jobspec/parse_group: use hcl decode not mapstructure, mount_options
* client/allocrunner/csi_hook: pass MountOptions through
client/allocrunner/csi_hook: add a VolumeMountOptions
client/allocrunner/csi_hook: drop Options
client/allocrunner/csi_hook: use the structs options
* client/pluginmanager/csimanager/interface: UsageOptions.MountOptions
* client/pluginmanager/csimanager/volume: pass MountOptions in capabilities
* plugins/csi/plugin: remove todo 7007 comment
* nomad/structs/csi: MountOptions
* api/csi: add options to the api for parsing, match structs
* plugins/csi/plugin: move VolumeMountOptions to structs
* api/csi: use specific type for mount_options
* client/allocrunner/csi_hook: merge MountOptions here
* rename CSIOptions to CSIMountOptions
* client/allocrunner/csi_hook
* client/pluginmanager/csimanager/volume
* nomad/structs/csi
* plugins/csi/fake/client: add PrevVolumeCapability
* plugins/csi/plugin
* client/pluginmanager/csimanager/volume_test: remove debugging
* client/pluginmanager/csimanager/volume: fix odd merging logic
* api: rename CSIOptions -> CSIMountOptions
* nomad/csi_endpoint: remove a 7007 comment
* command/alloc_status: show mount options in the volume list
* nomad/structs/csi: include MountOptions in the volume stub
* api/csi: add MountOptions to stub
* command/volume_status_csi: clean up csiVolMountOption, add it
* command/alloc_status: csiVolMountOption lives in volume_csi_status
* command/node_status: display mount flags
* nomad/structs/volumes: npe
* plugins/csi/plugin: npe in ToCSIRepresentation
* jobspec/parse_test: expand volume parse test cases
* command/agent/job_endpoint: ApiTgToStructsTG needs MountOptions
* command/volume_status_csi: copy paste error
* jobspec/test-fixtures/basic: hclfmt
* command/volume_status_csi: clean up csiVolMountOption
Run the plugin fingerprint one last time with a closed client during
instance manager shutdown. This will return quickly and will give us a
correctly-populated `PluginInfo` marked as unhealthy so the Nomad
client can update the server about plugin health.
Allow for faster updates to plugin status when allocations become
terminal by listening for register/deregister events from the dynamic
plugin registry (which in turn are triggered by the plugin supervisor
hook).
The deregistration function closures that we pass up to the CSI plugin
manager don't properly close over the name and type of the
registration, causing monolith-type plugins to deregister only one of
their two plugins on alloc shutdown. Rebind plugin supervisor
deregistration targets to fix that.
Includes log message and comment improvements
This changeset implements the remaining controller detach RPCs: server-to-client and client-to-controller. The tests also uncovered a bug in our RPC for claims which is fixed here; the volume claim RPC is used for both claiming and releasing a claim on a volume. We should only submit a controller publish RPC when the claim is new and not when it's being released.
In order to correctly fingerprint dynamic plugins on client restarts,
we need to persist a handle to the plugin (that is, connection info)
to the client state store.
The dynamic registry will sync automatically to the client state
whenever it receives a register/deregister call.
* nomad/structs/csi: new RemoteID() uses the ExternalID if set
* nomad/csi_endpoint: pass RemoteID to volume request types
* client/pluginmanager/csimanager/volume: pass RemoteID to NodePublishVolume
Fix some docstring typos and fix noisy log message during client restarts.
A log for the common case where the plugin socket isn't ready yet
isn't actionable by the operator so having it at info is just noise.
* client/allocrunner/csi_hook: tag errors
* nomad/client_csi_endpoint: tag errors
* nomad/client_rpc: remove an unnecessary error tag
* nomad/state/state_store: ControllerRequired fix intent
We use ControllerRequired to indicate that a volume should use the
publish/unpublish workflow, rather than that it has a controller. We
need to check both RequiresControllerPlugin and SupportsAttachDetach
from the fingerprint to check that.
* nomad/csi_endpoint: tag errors
* nomad/csi_endpoint_test: longer error messages, mock fingerprints
Derive a provider name and version for plugins (and the volumes that
use them) from the CSI identity API `GetPluginInfo`. Expose the vendor
name as `Provider` in the API and CLI commands.
* structs: CSIInfo include AllocID, CSIPlugins no Jobs
* state_store: eliminate plugin Jobs, delete an empty plugin
* nomad/structs/csi: detect empty plugins correctly
* client/allocrunner/taskrunner/plugin_supervisor_hook: option AllocID
* client/pluginmanager/csimanager/instance: allocID
* client/pluginmanager/csimanager/fingerprint: set AllocID
* client/node_updater: split controller and node plugins
* api/csi: remove Jobs
The CSI Plugin API will map plugins to allocations, which allows
plugins to be defined by jobs in many configurations. In particular,
multiple plugins can be defined in the same job, and multiple jobs can
be used to define a single plugin.
Because we now map the allocation context directly from the node, it's
no longer necessary to track the jobs associated with a plugin
directly.
* nomad/csi_endpoint_test: CreateTestPlugin & register via fingerprint
* client/dynamicplugins: lift AllocID into the struct from Options
* api/csi_test: remove Jobs test
* nomad/structs/csi: CSIPlugins has an array of allocs
* nomad/state/state_store: implement CSIPluginDenormalize
* nomad/state/state_store: CSIPluginDenormalize npe on missing alloc
* nomad/csi_endpoint_test: defer deleteNodes for clarity
* api/csi_test: disable this test awaiting mocks:
https://github.com/hashicorp/nomad/issues/7123
Currently the handling of CSINode RPCs does not correctly handle
forwarding RPCs to Nodes.
This commit fixes this by introducing a shim RPC
(nomad/client_csi_enpdoint) that will correctly forward the request to
the owning node, or submit the RPC to the client.
In the process it also cleans up handling a little bit by adding the
`CSIControllerQuery` embeded struct for required forwarding state.
The CSIControllerQuery embeding the requirement of a `PluginID` also
means we could move node targetting into the shim RPC if wanted in the
future.
CSI Plugins that manage devices need not just access to the CSI
directory, but also to manage devices inside `/dev`.
This commit introduces a `/dev:/dev` mount to the container so that they
may do so.
This commit is the initial implementation of claiming volumes from the
server and passes through any publishContext information as appropriate.
There's nothing too fancy here.
