Executes the provided actions on the JVM's main thread. Note that this is, by definition, a single-threaded executor, and should not be used for anything which requires a meaningful amount of performance. Additionally, and also by definition, this process conflicts with producing the results of an application. If one fiber calls evalOn(MainThread) while the main fiber is returning, the first one will "win" and will cause the second one to wait its turn. Once the main fiber produces results (or errors, or cancels), any remaining enqueued actions are ignored and discarded (a mostly irrelevant issue since the process is, at that point, terminating).

This is not recommended for use in most applications, and is really only appropriate for scenarios where some third-party library is sensitive to the exact identity of the calling thread (for example, LWJGL). In these scenarios, it is recommended that the absolute minimum possible amount of work is handed off to the main thread.

Configures whether to enable blocked thread detection. This is relatively expensive so is off by default and probably not something that you want to permanently enable in production.

If enabled, the compute pool will attempt to detect when blocking operations have been erroneously wrapped in IO.apply or IO.delay instead of IO.blocking or IO.interruptible and will report stacktraces of this to stderr.

This may be of interest if you've been getting warnings about CPU starvation printed to stderr. https://typelevel.org/cats-effect/docs/core/starvation-and-tuning

Can also be configured by setting the cats.effect.detectBlockedThreads system property.

Controls the number of worker threads which will be allocated to the compute pool in the underlying runtime. In general, this should be no greater than the number of physical threads made available by the underlying kernel (which can be determined using Runtime.getRuntime().availableProcessors()). For any application which has significant additional non-compute thread utilization (such as asynchronous I/O worker threads), it may be optimal to reduce the number of compute threads by the corresponding amount such that the total number of active threads exactly matches the number of underlying physical threads.

In practice, tuning this parameter is unlikely to affect your application performance beyond a few percentage points, and the default value is optimal (or close to optimal) in most common scenarios.

This setting is JVM-specific and will not compile on JavaScript.

For more details on Cats Effect's runtime threading model please see https://typelevel.org/cats-effect/docs/thread-model.

Controls whether non-daemon threads blocking application exit are logged to stderr when the IO produced by run has completed. This mechanism works by starting a daemon thread which periodically polls all active threads on the system, checking for any remaining non-daemon threads and enumerating them. This can be very useful for determining why your application isn't gracefully exiting, since the alternative is that the JVM will just hang waiting for the non-daemon threads to terminate themselves. This mechanism will not, by itself, block shutdown in any way. For this reason, it defaults to true.

In the event that your application exit is being blocked by a non-daemon thread which you cannot control (i.e. a bug in some dependency), you can circumvent the blockage by appending the following to the IO returned from run:

val program: IO[ExitCode] = ???                      // the original IO returned from `run`
program.guarantee(IO(Runtime.getRuntime().halt(0)))  // the bit you need to add

This finalizer will forcibly terminate the JVM (kind of like kill -9), ignoring daemon threads and shutdown hooks, but only after all native Cats Effect finalizers have completed. In most cases, this should be a relatively benign thing to do, though it's definitely a bad default. Only use this to workaround a blocking non-daemon thread that you cannot otherwise influence!

Can also be configured by setting the cats.effect.logNonDaemonThreadsOnExit system property.

Controls the interval used by the non-daemon thread detector. Defaults to 10.seconds.

Can also be configured by setting the cats.effect.logNonDaemonThreads.sleepIntervalMillis system property.

Defines what to do when CpuStarvationCheck is triggered. Defaults to log a warning to System.err.

Configures the action to perform when unhandled errors are caught by the runtime. An unhandled error is an error that is raised (and not handled) on a Fiber that nobody is joining.

For example:

import scala.concurrent.duration._
override def run: IO[Unit] = IO(throw new Exception("")).start *> IO.sleep(1.second)

In this case, the exception is raised on a Fiber with no listeners. Nobody would be notified about that error. Therefore it is unhandled, and it goes through the reportFailure mechanism.

By default, reportFailure simply delegates to cats.effect.std.Console!.printStackTrace. It is safe to perform any IO action within this handler; it will not block the progress of the runtime. With that said, some care should be taken to avoid raising unhandled errors as a result of handling unhandled errors, since that will result in the obvious chaos.

The entry point for your application. Will be called by the runtime when the process is started. If the underlying runtime supports it, any arguments passed to the process will be made available in the args parameter. The numeric value within the resulting ExitCode will be used as the exit code when the process terminates unless terminated exceptionally or by interrupt.

The runtime which will be used by IOApp to evaluate the IO produced by the run method. This may be overridden by IOApp implementations which have extremely specialized needs, but this is highly unlikely to ever be truly needed. As an example, if an application wishes to make use of an alternative compute thread pool (such as Executors.fixedThreadPool), it is almost always better to leverage IO.evalOn on the value produced by the run method, rather than directly overriding runtime.

In other words, this method is made available to users, but its use is strongly discouraged in favor of other, more precise solutions to specific use-cases.

This value is guaranteed to be equal to unsafe.IORuntime.global.

The configuration used to initialize the runtime which will evaluate the IO produced by run. It is very unlikely that users will need to override this method.