Add flexible scaling patterns to FAQ (#491)

* Add flexible scaling patterns to FAQ

* Final edits

Author: shikokuchuo

dev/vignettes/_v07-questions.Rmd

@@ -128,3 +128,37 @@ func <- carrier::crate(\(x) fn(x), fn = fn)
 ```
 
 - For environments, consider using `parent.env(e) <- emptyenv()`. This is not required for R6 classes as they are already isolated by default. Regardless of any parent environment, an environment / R6 class could still contain many items that are not needed by the parallel process, in which case consider passing individual members (`env$x`, `env$y`) rather than the entire object (`env`) where possible.
+
+### 5. How to create daemons on-demand or shutdown daemons when they're not used
+
+Setting daemons can be totally separate from the task of actually launching (deploying) them. To just set daemons when you're using only your local machine:
+
+```r
+daemons(url = local_url())
+```
+or if you're using your local machine and/or remote machines:
+
+```r
+daemons(url = host_url())
+```
+This sets up daemons. You can think of this as a 'base station' that listens for incoming daemon connections.
+
+When you actually want to launch (deploy) a daemon, you may use either:
+```r
+launch_local()
+```
+or
+```r
+launch_remote(remote = ssh_config("ssh://servername")) # or cluster_config()
+```
+The key to adopting a flexible-use pattern that lets you scale up and down is to specify one of the following arguments to the `...` argument of `launch_local()` or `launch_remote()`. You may also supply these to the initial `daemons()` call for them to apply by default for all launches:
+
+- `maxtasks`: integer number of tasks to perform before exiting.
+- `idletime`: milliseconds idle time before exiting.
+- `walltime`: milliseconds (soft) wall time before exiting (i.e. at least this amount of time, and possibly more as there is no forcible timeout mid-task)
+
+As an example, to launch a daemon to perform one task only:
+```r
+launch_remote(remote = ssh_config("ssh://servername"), maxtasks = 1L)
+```
+In this way, you may use `cluster_config()` to launch jobs on an HPC cluster on demand, without using persistent daemons. Note that you will incur the latency costs of however long it takes to launch the job.

vignettes/v07-questions.Rmd

@@ -136,3 +136,37 @@ func <- carrier::crate(\(x) fn(x), fn = fn)
 ```
 
 - For environments, consider using `parent.env(e) <- emptyenv()`. This is not required for R6 classes as they are already isolated by default. Regardless of any parent environment, an environment / R6 class could still contain many items that are not needed by the parallel process, in which case consider passing individual members (`env$x`, `env$y`) rather than the entire object (`env`) where possible.
+
+### 5. How to create daemons on-demand or shutdown daemons when they're not used
+
+Setting daemons can be totally separate from the task of actually launching (deploying) them. To just set daemons when you're using only your local machine:
+
+```r
+daemons(url = local_url())
+```
+or if you're using your local machine and/or remote machines:
+
+```r
+daemons(url = host_url())
+```
+This sets up daemons. You can think of this as a 'base station' that listens for incoming daemon connections.
+
+When you actually want to launch (deploy) a daemon, you may use either:
+```r
+launch_local()
+```
+or
+```r
+launch_remote(remote = ssh_config("ssh://servername")) # or cluster_config()
+```
+The key to adopting a flexible-use pattern that lets you scale up and down is to specify one of the following arguments to the `...` argument of `launch_local()` or `launch_remote()`. You may also supply these to the initial `daemons()` call for them to apply by default for all launches:
+
+- `maxtasks`: integer number of tasks to perform before exiting.
+- `idletime`: milliseconds idle time before exiting.
+- `walltime`: milliseconds (soft) wall time before exiting (i.e. at least this amount of time, and possibly more as there is no forcible timeout mid-task)
+
+As an example, to launch a daemon to perform one task only:
+```r
+launch_remote(remote = ssh_config("ssh://servername"), maxtasks = 1L)
+```
+In this way, you may use `cluster_config()` to launch jobs on an HPC cluster on demand, without using persistent daemons. Note that you will incur the latency costs of however long it takes to launch the job.