1 /*--------------------------------------------------------------------------
4 * Code for sample worker making use of shared memory message queues.
5 * Our test worker simply reads messages from one message queue and
6 * writes them back out to another message queue. In a real
7 * application, you'd presumably want the worker to do some more
8 * complex calculation rather than simply returning the input,
9 * but it should be possible to use much of the control logic just
12 * Copyright (c) 2013-2022, PostgreSQL Global Development Group
15 * src/test/modules/test_shm_mq/worker.c
17 * -------------------------------------------------------------------------
22 #include "miscadmin.h"
23 #include "storage/ipc.h"
24 #include "storage/procarray.h"
25 #include "storage/shm_mq.h"
26 #include "storage/shm_toc.h"
27 #include "tcop/tcopprot.h"
29 #include "test_shm_mq.h"
31 static void attach_to_queues(dsm_segment
*seg
, shm_toc
*toc
,
32 int myworkernumber
, shm_mq_handle
**inqhp
,
33 shm_mq_handle
**outqhp
);
34 static void copy_messages(shm_mq_handle
*inqh
, shm_mq_handle
*outqh
);
37 * Background worker entrypoint.
39 * This is intended to demonstrate how a background worker can be used to
40 * facilitate a parallel computation. Most of the logic here is fairly
41 * boilerplate stuff, designed to attach to the shared memory segment,
42 * notify the user backend that we're alive, and so on. The
43 * application-specific bits of logic that you'd replace for your own worker
44 * are attach_to_queues() and copy_messages().
47 test_shm_mq_main(Datum main_arg
)
53 volatile test_shm_mq_header
*hdr
;
58 * Establish signal handlers.
60 * We want CHECK_FOR_INTERRUPTS() to kill off this worker process just as
61 * it would a normal user backend. To make that happen, we use die().
63 pqsignal(SIGTERM
, die
);
64 BackgroundWorkerUnblockSignals();
67 * Connect to the dynamic shared memory segment.
69 * The backend that registered this worker passed us the ID of a shared
70 * memory segment to which we must attach for further instructions. Once
71 * we've mapped the segment in our address space, attach to the table of
72 * contents so we can locate the various data structures we'll need to
73 * find within the segment.
75 * Note: at this point, we have not created any ResourceOwner in this
76 * process. This will result in our DSM mapping surviving until process
77 * exit, which is fine. If there were a ResourceOwner, it would acquire
78 * ownership of the mapping, but we have no need for that.
80 seg
= dsm_attach(DatumGetInt32(main_arg
));
83 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE
),
84 errmsg("unable to map dynamic shared memory segment")));
85 toc
= shm_toc_attach(PG_TEST_SHM_MQ_MAGIC
, dsm_segment_address(seg
));
88 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE
),
89 errmsg("bad magic number in dynamic shared memory segment")));
92 * Acquire a worker number.
94 * By convention, the process registering this background worker should
95 * have stored the control structure at key 0. We look up that key to
96 * find it. Our worker number gives our identity: there may be just one
97 * worker involved in this parallel operation, or there may be many.
99 hdr
= shm_toc_lookup(toc
, 0, false);
100 SpinLockAcquire(&hdr
->mutex
);
101 myworkernumber
= ++hdr
->workers_attached
;
102 SpinLockRelease(&hdr
->mutex
);
103 if (myworkernumber
> hdr
->workers_total
)
105 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE
),
106 errmsg("too many message queue testing workers already")));
109 * Attach to the appropriate message queues.
111 attach_to_queues(seg
, toc
, myworkernumber
, &inqh
, &outqh
);
114 * Indicate that we're fully initialized and ready to begin the main part
115 * of the parallel operation.
117 * Once we signal that we're ready, the user backend is entitled to assume
118 * that our on_dsm_detach callbacks will fire before we disconnect from
119 * the shared memory segment and exit. Generally, that means we must have
120 * attached to all relevant dynamic shared memory data structures by now.
122 SpinLockAcquire(&hdr
->mutex
);
123 ++hdr
->workers_ready
;
124 SpinLockRelease(&hdr
->mutex
);
125 registrant
= BackendPidGetProc(MyBgworkerEntry
->bgw_notify_pid
);
126 if (registrant
== NULL
)
128 elog(DEBUG1
, "registrant backend has exited prematurely");
131 SetLatch(®istrant
->procLatch
);
134 copy_messages(inqh
, outqh
);
137 * We're done. For cleanliness, explicitly detach from the shared memory
138 * segment (that would happen anyway during process exit, though).
145 * Attach to shared memory message queues.
147 * We use our worker number to determine to which queue we should attach.
148 * The queues are registered at keys 1..<number-of-workers>. The user backend
149 * writes to queue #1 and reads from queue #<number-of-workers>; each worker
150 * reads from the queue whose number is equal to its worker number and writes
151 * to the next higher-numbered queue.
154 attach_to_queues(dsm_segment
*seg
, shm_toc
*toc
, int myworkernumber
,
155 shm_mq_handle
**inqhp
, shm_mq_handle
**outqhp
)
160 inq
= shm_toc_lookup(toc
, myworkernumber
, false);
161 shm_mq_set_receiver(inq
, MyProc
);
162 *inqhp
= shm_mq_attach(inq
, seg
, NULL
);
163 outq
= shm_toc_lookup(toc
, myworkernumber
+ 1, false);
164 shm_mq_set_sender(outq
, MyProc
);
165 *outqhp
= shm_mq_attach(outq
, seg
, NULL
);
169 * Loop, receiving and sending messages, until the connection is broken.
171 * This is the "real work" performed by this worker process. Everything that
172 * happens before this is initialization of one form or another, and everything
173 * after this point is cleanup.
176 copy_messages(shm_mq_handle
*inqh
, shm_mq_handle
*outqh
)
184 /* Notice any interrupts that have occurred. */
185 CHECK_FOR_INTERRUPTS();
187 /* Receive a message. */
188 res
= shm_mq_receive(inqh
, &len
, &data
, false);
189 if (res
!= SHM_MQ_SUCCESS
)
192 /* Send it back out. */
193 res
= shm_mq_send(outqh
, len
, data
, false, true);
194 if (res
!= SHM_MQ_SUCCESS
)