Class BatchClusterOptions

These parameter values have somewhat sensible defaults, but can be overridden for a given BatchCluster.

Constructors

Properties

cleanupChildProcs: boolean = true

Should batch-cluster try to clean up after spawned processes that don't shut down?

Only disable this if you have another means of PID cleanup.

Defaults to true.

endGracefulWaitTimeMillis: number = 500

When this.end() is called, or Node broadcasts the beforeExit event, this is the milliseconds spent waiting for currently running tasks to finish before sending kill signals to child processes.

Setting this value to 0 means child processes will immediately receive a kill signal to shut down. Any pending requests may be interrupted. Must be >= 0. Defaults to 500ms.

healthCheckIntervalMillis: number = 0

If healthCheckCommand is set, how frequently should we check for unhealthy child processes?

Set this to 0 to disable this feature.

logger: (() => Logger) = logger

A BatchCluster instance and associated BatchProcess instances will share this Logger. Defaults to the Logger instance provided to setLogger().

Type declaration

maxFailedTasksPerProcess: number = 2

How many failed tasks should a process be allowed to process before it is recycled?

Set this to 0 to disable this feature.

maxIdleMsPerProcess: number = 0

If a child process is idle for more than this value (in milliseconds), shut it down to reduce system resource consumption.

A value of ~10 seconds to a couple minutes would be reasonable. Set this to 0 to disable this feature.

maxProcAgeMillis: number = ...

Child processes will be recycled when they reach this age.

This value must not be less than spawnTimeoutMillis or taskTimeoutMillis.

Defaults to 5 minutes. Set to 0 to disable.

maxProcs: number = 1

No more than maxProcs child processes will be run at a given time to serve pending tasks.

Defaults to 1.

maxReasonableProcessFailuresPerMinute: number = 10

If the initial versionCommand fails for new spawned processes more than this rate, end this BatchCluster and throw an error, because something is terribly wrong.

If this backstop didn't exist, new (failing) child processes would be created indefinitely.

Defaults to 10. Set to 0 to disable.

maxTasksPerProcess: number = 500

Processes will be recycled after processing maxTasksPerProcess tasks. Depending on the commands and platform, batch mode commands shouldn't exhibit unduly memory leaks for at least tens if not hundreds of tasks. Setting this to a low number (like less than 10) will impact performance markedly, due to OS process start/stop maintenance. Setting this to a very high number (> 1000) may result in more memory being consumed than necessary.

Must be >= 0. Defaults to 500

minDelayBetweenSpawnMillis: number = ...

If maxProcs > 1, spawning new child processes to process tasks can slow down initial processing, and create unnecessary processes.

Must be >= 0ms. Defaults to 1.5 seconds.

onIdleIntervalMillis: number = ...

This is the minimum interval between calls to BatchCluster's #onIdle method, which runs general janitorial processes like child process management and task queue validation.

Must be > 0. Defaults to 10 seconds.

pidCheckIntervalMillis: number = ...

Verify child processes are still running by checking the OS process table.

Set this to 0 to disable this feature.

spawnTimeoutMillis: number = ...

Spawning new child processes and servicing a "version" task must not take longer than spawnTimeoutMillis before the process is considered failed, and need to be restarted. Be pessimistic here--windows can regularly take several seconds to spin up a process, thanks to antivirus shenanigans.

Defaults to 15 seconds. Set to 0 to disable.

streamFlushMillis: number = ...

When a task sees a "pass" or "fail" from either stdout or stderr, it needs to wait for the other stream to finish flushing to ensure the task's Parser sees the entire relevant stream contents. A larger number may be required for slower computers to prevent internal errors due to lack of stream coercion.

Note that this puts a hard lower limit on task latency, so don't set this to a large number: no task will resolve faster than this value (in millis).

If you set this value too low, tasks may be erroneously resolved or rejected (depending on which stream is handled first).

Your system may support a smaller value: this is a pessimistic default. If this is set too low, you'll see noTaskData events.

Setting this to 0 makes whatever flushes first--stdout and stderr--and will most likely result in internal errors (due to stream buffers not being able to be associated to tasks that were just settled)

taskTimeoutMillis: number = ...

If commands take longer than this, presume the underlying process is dead and we should fail the task.

This should be set to something on the order of seconds.

Defaults to 10 seconds. Set to 0 to disable.

Generated using TypeDoc