exit

The format of the base version is:

/admin?op=exit

The exit operation works as follows:

• Any new non-admin request will get an HTTP response code 503 (Service Unavailable).
• Any in-progress request will finish normally (including updates).
• The MDEX Engine will wait to exit until the following conditions have been met:
  • All requests have finished (including updates).
  • All background merging has been completed.

The exit operation's output to the browser looks similar to the following:

Dgraph admin, OK
Dgraph shutting down at Thu Feb 17 13:12:54 2011

The command also writes shutdown information to the Dgraph error log, as in this example:

Shutdown request with received at Thu Feb 17 13:12:54 2011.
Shutdown will complete when all outstanding jobs are complete.
All dgraph transactions completed at Thu Feb 17 13:12:54 2011, exiting normally (pid=4128)

The timeout version lets you specify a time limit, in seconds, of the shutdown procedure. The format of the timeout version is:

/admin?op=exit&timelimit=seconds

where seconds is a positive integer.

This example uses a time limit of 30 seconds:

/admin?op=exit&timelimit=30

The exit&timelimit operation works as follows:

• A time limit of 0 (zero) will shut down the MDEX Engine immediately.
• Any queries still in progress when the time limit is reached will not return a result to the client (i.e., the client will observe a closed connection).
• Any queries still in progress when the time limit is reached will not be logged in the Dgraph log.
• Any updates still in progress when the time limit is reached will not be applied.
• Any background merges still in progress when the time limit is reached will be aborted at the end of the timeout.
• The number of in-progress queries is written to the Dgraph error log just before exiting, along with a message stating that the shutdown time limit was reached.

Issuing an exit command with a time limit ensures that the MDEX Engine shuts down within that time limit, regardless of prior or following exit queries (i.e., exit commands with a time limit can only shorten the MDEX Engine's time to live). These examples demonstrate what happens when successive exit commands are issued:

• If exit&timelimit=30 is issued and 10 seconds later exit&timelimit=0 is issued, the MDEX Engine will exit immediately when the second request is issued (if it hasn't already exited).
• If exit&timelimit=30 is issued and 10 seconds later exit&timelimit=5 is issued, the MDEX Engine will exit 5 seconds after the second request, or when all queries are drained (whichever comes first).
• If exit&timelimit=30 is issued and 10 seconds later exit&timelimit=30 is issued, the MDEX Engine will exit 30 seconds after the first request, or when all queries are drained (whichever comes first).
• If exit&timelimit=30 is issued and 10 seconds later the base exit command is issued, the MDEX Engine will exit 30 seconds after the first request, or when all queries are drained (whichever comes first).
• If exit is issued and 10 seconds later exit&timelimit=30 is issued, the MDEX Engine will exit 30 seconds after the second request, or when all queries are drained (whichever comes first).

An exit&timelimit=30 operation's output to the browser looks like this:

Dgraph admin, OK
Dgraph shutting down within 30 seconds at Thu Feb 17 14:09:38 2011 

The command also writes shutdown information to the Dgraph error log, as in this example for an exit&timelimit=30 command:

Shutdown request with time limit of 30 seconds received at Thu Feb 17 14:09:38 2011.
Shutdown will complete when all outstanding jobs are complete, or within 30 seconds, whichever happens earlier.
All dgraph transactions completed at Thu Feb 17 13:12:54 2011, exiting normally (pid=4128)

If command were for an immediate shutdown (exit&timelimit=0) and queries were still in progress, the Dgraph error log would contain a message similar to this example:

Shutdown request with time limit of 0 seconds received at Thu Feb 17 14:18:46 2011.
Shutdown will complete when all outstanding jobs are complete, or within 0 seconds, whichever happens earlier.
Shutdown time limit reached at Thu Feb 17 14:18:46 2011, exiting with jobs still in progress:
1 request is still active or queued, and will not be logged to the request log
1 job is currently executing, and will be killed.
0 jobs are queued, and will not be executed.