The controller learns IP addresses from agents that peer with it, but that
information has so far gotten lost when resulting configs get pushed out to the
agents. This makes these updates include that information.
This is quite redundant with the enumeration for Broker ports,
unfortunately. But the logic is subtly different: all nodes obtain a telemetry
port, while not all nodes require a Broker port, for example, and in the metrics
port assignment we also cross-check selected Broker ports. I found more unified
code actually harder to read in the end.
The logic for the two sets remains the same: from a start point, ports get
enumerated sequentially that aren't otherwise taken. These ports are assumed
available; there's nothing that checks their availability -- for now.
The default start port is 9000. I considered 9090, to align with the Prometheus
default, but counting upward from there is likely to hit trouble with the Broker
default ports (9999/9997), used by the Supervisor. Counting downward is a bit
unnatural, and shifting the Broker default ports brings subtle ordering issues.
This also changes the node ordering logic slightly since it seems more intuitive
to keep sequential ports on a given instance, instead of striping across them.
This eliminates one place in which we currently need to mirror changes to the
script-land Cluster::Node record. Instead of keeping an exact in-core equivalent, the
Supervisor now treats the data structure as opaque, and stores the whole cluster
table as a JSON string.
We may replace the script-layer Supervisor::ClusterEndpoint in the future, using
Cluster::Node directly. But that's a more invasive change that will affect how
people invoke Supervisor::create() and similars.
Relying on JSON for serialization has the side-effect of removing the
Supervisor's earlier quirk of using 0/tcp, not 0/unknown, to indicate unused
ports in the Supervisor::ClusterEndpoint record.
The controller now listens on an additional port, defaulting to 2149, for Broker
connections via websockets. Configuration works as for the existing traditional
Broker port (2150), via ZEEK_CONTROLLER_WEBSOCKET_ADDR and
ZEEK_CONTROLLER_WEBSOCKET_PORT environment variables, as well as corresponding
redef'able constants.
To disable the websockets feature, leave ZEEK_CONTROLLER_WEBSOCKET_PORT unset
and redefine Management::Controller::default_port_websocket to 0/unknown.
This erroneously used connectedness of instances, not presence of a deployed
cluster. Without a deployment, there's no point in trying to retrieve global ID
values.
Ensure the framework's log stream exists prior to using it in zeek_init(), and
use a node-is-live message similar to those in agent and controller also in
launched nodes.
Failing to do so could open a race condition in which a quickly connecting
controller could send instructions whose resulting Supervisor interactions got
lost.
This adds restart request/response event pairs that restart nodes in the running
Zeek cluster. The implementation is very similar to get_id_value, which also
involves distributing a list of nodes to agents and aggregating the responses.
This declares our helper functions for sending events to the Supervisor, and
makes them return the created request objects to enable the caller to modify
them. It also adds a helper for restart and status requests, uses the helpers
throughout the module, and makes all handlers more resilient in case Supervisor
events other than the agent's arrive.
The controller now logs its deployment attempt of a persisted configuration at
startup. This is generally helpful to see recorded, and also explains timeout of
the underlying request in case of failure (which triggers a timeout message).
For the case of a running cluster with no connected agents, use the
g_instances_known table instead of g_instances. The latter reflects the contents
of the last deployed config, not the live scenario of actually attached agents.
The timeout result wasn't actually stored in requests timing out in the
agent. (So far that's for deployment requests.) Also log the timing out of any
request state, similar to the controller.
No functional change, just a consistency tweak. Since agent and controller send
response events via Broker::publish(), the arguments aren't named and so this
only affects the API definition.
More resilience: when an agent restarts, it checks in with the controller. If
the controller has deployed a config, this check-in may lead to an internal
notify_agents_ready event. At that point, we now trigger a deployment when there
currently isn't already one running. This ensures that any agents not yet
running the current cluster will start to do so, and does nothing when those
agents already run it, since they ignore the request in that case.
When agents had to terminate existing Zeek cluster nodes at the beginning of a
new deployment, they so far used their internal state to look up the nodes and
fired off requests to the Supervisor to shut these down. This has a problem:
when an agent restarts unexpectedly, it has no internal state, and when it then
tries to create nodes that already exist, the Supervisor complains with error
messages.
To avoid this, the agent now tears down all Supervised nodes other than agents
and controllers. In order to do so, it first needs to query the Supervisor for
the current node status, which means there are now two such status requests: one
upon deployment, and one during get_nodes requests. In order to disambiguate
these contexts in the SupervisorControl::status_request/response transactions,
we use the finish() callback in the corresponding request state to continue
execution as needed.
A resilience feature: when a booting controller has a previously deployed
configuration (just reloaded from persistent state), it now triggers a
deployment. When agents at this point run something else, this restores the
controller's understanding of what's deployed, and if the agents do still run
this configuration, does nothing since agents ignore deployment of a
configuration they already run.
The controller now runs most of a config deployment via an internal function,
allowing it to be called from multiple places instead of just the deploy_request
event handler.
The controller's deployment request state now features a bit that indicates
whether the deployment was requested by a client, or triggered internally. This
affects logging and the transmission of deployment response events via Broker,
which are skipped when the deployment is internal.
This is in preparation of resilience features when the controller (re-)boots.
This allows us to handle loss of Broker peerings, updating instance state as we
see instances go away. This also tweaks logging slightly to differentiate
between an instance checking in for the first time, and checking in when the
controller already knows it.
These callbacks are handy for stringing together codepaths separated by event
request/response transactions: when such a transaction completes, the callback
allows locating a parent request for the finished one, to continue its
processing.
When an agent is already running the configuration it's asked to deploy,
it will now recognize this and by default do nothing. The requester can force
it if needed, via a new argument to the deploy_request event.
The agent's Broker::peer_added handler now recognizes the Supervisor and does
not trigger a notify_agent_hello event upon it. It might still send such events
repeatedly as other things peer with the agent.
The controller now knows three states that a cluster configuration can be in:
- STAGED: as uploaded by the client
- READY: with needed tweaks applied, e.g. to fill in ports
- DEPLOYED: as sent off to agents for deployment
These states aren't exclusive, they represent checkpoints that a config goes
through from upload through deployment. A deployed configuration will also exist
in its STAGED and READY versions, unless a client has uploaded a new
configuration, which will overwrite the STAGED and READY ones.
The controller saves all of these in a table, which lets us use Broker to
persist all states to disk. We use &broker_allow_complex_type, since we only
ever store entire configurations.
This separates uploading a configuration from deploying it to the instances into
separate event transactions. set_configuration_request/response remains, but now
only conducts validation and storage of the new configuration (upon validation
success, and not yet persisted to disk). The response event indicates success or
the list of validation errors. Successful upload now returns the configuration's
ID in the result record's data struct.
The new deploy_request/response event takes a previously uploaded configuration
and deploys it to the agents.
The controller now tracks uploaded and deployed configurations
separately. Uploading assigns g_config_staged; deployment assigns
g_config_deployed. Deployment does not affect g_config_staged.
The get_config_request/response event pair now allows selecting the
configuration the caller would like to retrieve.
This renames the agent's functionality for setting a configuration to reflect
the controller's upcoming separation of set_configuration and deployment.
The instance and error fields are now optional instead of defaulting to empty
strings, which caused minor output deviations in the client.
Agents now ensure that any Result record they create has the instance field
filled in.
During `set_configuration_request` handling the controller now validates
received configurations, checking for a few common gotchas around naming and
port use. Validation continues once it finds a problem, resulting in a list
summarizing all identified problems.
The numbering process now accounts for the possibility of colliding with the
agent port, as well as with ports explicitly assigned in the configuration. It
also avoids nondeterminism that could result from traversal of sets.
It helps during testing to be able to control whether the Supervisor process
also routs node output to the console, in addition to writing to output
files. Since the Supervisor runs as the main process in Docker containers, its
output becomes visible in "docker logs" that way, simplifying diagnostics.
When the controller receives a configuration with no instances (and thus no
nodes), it needs to roundtrip to agents and can send the response right away.
This makes agents handle log archival automatically. By default, they invoke
zeek-archiver once every log rotation interval to archive rotated files from the
log-queue spool directory into the installation's log directory. The user can
disable the feature, customize the command to invoke, and adjust the rotation
interval.
