Branch data Line data Source code
1 : : /*--------------------------------------------------------------------------
2 : : *
3 : : * setup.c
4 : : * Code to set up a dynamic shared memory segments and a specified
5 : : * number of background workers for shared memory message queue
6 : : * testing.
7 : : *
8 : : * Copyright (c) 2013-2026, PostgreSQL Global Development Group
9 : : *
10 : : * IDENTIFICATION
11 : : * src/test/modules/test_shm_mq/setup.c
12 : : *
13 : : * -------------------------------------------------------------------------
14 : : */
15 : :
16 : : #include "postgres.h"
17 : :
18 : : #include "miscadmin.h"
19 : : #include "pgstat.h"
20 : : #include "postmaster/bgworker.h"
21 : : #include "storage/proc.h"
22 : : #include "storage/shm_toc.h"
23 : : #include "test_shm_mq.h"
24 : : #include "utils/memutils.h"
25 : : #include "utils/wait_event.h"
26 : :
27 : : typedef struct
28 : : {
29 : : int nworkers;
30 : : BackgroundWorkerHandle *handle[FLEXIBLE_ARRAY_MEMBER];
31 : : } worker_state;
32 : :
33 : : static void setup_dynamic_shared_memory(int64 queue_size, int nworkers,
34 : : dsm_segment **segp,
35 : : test_shm_mq_header **hdrp,
36 : : shm_mq **outp, shm_mq **inp);
37 : : static worker_state *setup_background_workers(int nworkers,
38 : : dsm_segment *seg);
39 : : static void cleanup_background_workers(dsm_segment *seg, Datum arg);
40 : : static void wait_for_workers_to_become_ready(worker_state *wstate,
41 : : volatile test_shm_mq_header *hdr);
42 : : static bool check_worker_status(worker_state *wstate);
43 : :
44 : : /* value cached, fetched from shared memory */
45 : : static uint32 we_bgworker_startup = 0;
46 : :
47 : : /*
48 : : * Set up a dynamic shared memory segment and zero or more background workers
49 : : * for a test run.
50 : : */
51 : : void
52 : 5 : test_shm_mq_setup(int64 queue_size, int32 nworkers, dsm_segment **segp,
53 : : shm_mq_handle **output, shm_mq_handle **input)
54 : : {
55 : : dsm_segment *seg;
56 : : test_shm_mq_header *hdr;
57 : 5 : shm_mq *outq = NULL; /* placate compiler */
58 : 5 : shm_mq *inq = NULL; /* placate compiler */
59 : : worker_state *wstate;
60 : :
61 : : /* Set up a dynamic shared memory segment. */
62 : 5 : setup_dynamic_shared_memory(queue_size, nworkers, &seg, &hdr, &outq, &inq);
63 : 5 : *segp = seg;
64 : :
65 : : /* Register background workers. */
66 : 5 : wstate = setup_background_workers(nworkers, seg);
67 : :
68 : : /* Attach the queues. */
69 : 5 : *output = shm_mq_attach(outq, seg, wstate->handle[0]);
70 : 5 : *input = shm_mq_attach(inq, seg, wstate->handle[nworkers - 1]);
71 : :
72 : : /* Wait for workers to become ready. */
73 : 5 : wait_for_workers_to_become_ready(wstate, hdr);
74 : :
75 : : /*
76 : : * Once we reach this point, all workers are ready. We no longer need to
77 : : * kill them if we die; they'll die on their own as the message queues
78 : : * shut down.
79 : : */
80 : 5 : cancel_on_dsm_detach(seg, cleanup_background_workers,
81 : : PointerGetDatum(wstate));
82 : 5 : pfree(wstate);
83 : 5 : }
84 : :
85 : : /*
86 : : * Set up a dynamic shared memory segment.
87 : : *
88 : : * We set up a small control region that contains only a test_shm_mq_header,
89 : : * plus one region per message queue. There are as many message queues as
90 : : * the number of workers, plus one.
91 : : */
92 : : static void
93 : 5 : setup_dynamic_shared_memory(int64 queue_size, int nworkers,
94 : : dsm_segment **segp, test_shm_mq_header **hdrp,
95 : : shm_mq **outp, shm_mq **inp)
96 : : {
97 : : shm_toc_estimator e;
98 : : int i;
99 : : Size segsize;
100 : : dsm_segment *seg;
101 : : shm_toc *toc;
102 : : test_shm_mq_header *hdr;
103 : :
104 : : /* Ensure a valid queue size. */
105 [ + - - + ]: 5 : if (queue_size < 0 || ((uint64) queue_size) < shm_mq_minimum_size)
106 [ # # ]: 0 : ereport(ERROR,
107 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
108 : : errmsg("queue size must be at least %zu bytes",
109 : : shm_mq_minimum_size)));
110 : : if (queue_size != ((Size) queue_size))
111 : : ereport(ERROR,
112 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
113 : : errmsg("queue size overflows size_t")));
114 : :
115 : : /*
116 : : * Estimate how much shared memory we need.
117 : : *
118 : : * Because the TOC machinery may choose to insert padding of oddly-sized
119 : : * requests, we must estimate each chunk separately.
120 : : *
121 : : * We need one key to register the location of the header, and we need
122 : : * nworkers + 1 keys to track the locations of the message queues.
123 : : */
124 : 5 : shm_toc_initialize_estimator(&e);
125 : 5 : shm_toc_estimate_chunk(&e, sizeof(test_shm_mq_header));
126 [ + + ]: 17 : for (i = 0; i <= nworkers; ++i)
127 : 12 : shm_toc_estimate_chunk(&e, (Size) queue_size);
128 : 5 : shm_toc_estimate_keys(&e, 2 + nworkers);
129 : 5 : segsize = shm_toc_estimate(&e);
130 : :
131 : : /* Create the shared memory segment and establish a table of contents. */
132 : 5 : seg = dsm_create(shm_toc_estimate(&e), 0);
133 : 5 : toc = shm_toc_create(PG_TEST_SHM_MQ_MAGIC, dsm_segment_address(seg),
134 : : segsize);
135 : :
136 : : /* Set up the header region. */
137 : 5 : hdr = shm_toc_allocate(toc, sizeof(test_shm_mq_header));
138 : 5 : SpinLockInit(&hdr->mutex);
139 : 5 : hdr->workers_total = nworkers;
140 : 5 : hdr->workers_attached = 0;
141 : 5 : hdr->workers_ready = 0;
142 : 5 : shm_toc_insert(toc, 0, hdr);
143 : :
144 : : /* Set up one message queue per worker, plus one. */
145 [ + + ]: 17 : for (i = 0; i <= nworkers; ++i)
146 : : {
147 : : shm_mq *mq;
148 : :
149 : 12 : mq = shm_mq_create(shm_toc_allocate(toc, (Size) queue_size),
150 : : (Size) queue_size);
151 : 12 : shm_toc_insert(toc, i + 1, mq);
152 : :
153 [ + + ]: 12 : if (i == 0)
154 : : {
155 : : /* We send messages to the first queue. */
156 : 5 : shm_mq_set_sender(mq, MyProc);
157 : 5 : *outp = mq;
158 : : }
159 [ + + ]: 12 : if (i == nworkers)
160 : : {
161 : : /* We receive messages from the last queue. */
162 : 5 : shm_mq_set_receiver(mq, MyProc);
163 : 5 : *inp = mq;
164 : : }
165 : : }
166 : :
167 : : /* Return results to caller. */
168 : 5 : *segp = seg;
169 : 5 : *hdrp = hdr;
170 : 5 : }
171 : :
172 : : /*
173 : : * Register background workers.
174 : : */
175 : : static worker_state *
176 : 5 : setup_background_workers(int nworkers, dsm_segment *seg)
177 : : {
178 : : MemoryContext oldcontext;
179 : : BackgroundWorker worker;
180 : : worker_state *wstate;
181 : : int i;
182 : :
183 : : /*
184 : : * We need the worker_state object and the background worker handles to
185 : : * which it points to be allocated in CurTransactionContext rather than
186 : : * ExprContext; otherwise, they'll be destroyed before the on_dsm_detach
187 : : * hooks run.
188 : : */
189 : 5 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
190 : :
191 : : /* Create worker state object. */
192 : 5 : wstate = MemoryContextAlloc(TopTransactionContext,
193 : 5 : offsetof(worker_state, handle) +
194 : : sizeof(BackgroundWorkerHandle *) * nworkers);
195 : 5 : wstate->nworkers = 0;
196 : :
197 : : /*
198 : : * Arrange to kill all the workers if we abort before all workers are
199 : : * finished hooking themselves up to the dynamic shared memory segment.
200 : : *
201 : : * If we die after all the workers have finished hooking themselves up to
202 : : * the dynamic shared memory segment, we'll mark the two queues to which
203 : : * we're directly connected as detached, and the worker(s) connected to
204 : : * those queues will exit, marking any other queues to which they are
205 : : * connected as detached. This will cause any as-yet-unaware workers
206 : : * connected to those queues to exit in their turn, and so on, until
207 : : * everybody exits.
208 : : *
209 : : * But suppose the workers which are supposed to connect to the queues to
210 : : * which we're directly attached exit due to some error before they
211 : : * actually attach the queues. The remaining workers will have no way of
212 : : * knowing this. From their perspective, they're still waiting for those
213 : : * workers to start, when in fact they've already died.
214 : : */
215 : 5 : on_dsm_detach(seg, cleanup_background_workers,
216 : : PointerGetDatum(wstate));
217 : :
218 : : /* Configure a worker. */
219 : 5 : memset(&worker, 0, sizeof(worker));
220 : 5 : worker.bgw_flags = BGWORKER_SHMEM_ACCESS;
221 : 5 : worker.bgw_start_time = BgWorkerStart_ConsistentState;
222 : 5 : worker.bgw_restart_time = BGW_NEVER_RESTART;
223 : 5 : sprintf(worker.bgw_library_name, "test_shm_mq");
224 : 5 : sprintf(worker.bgw_function_name, "test_shm_mq_main");
225 : 5 : snprintf(worker.bgw_type, BGW_MAXLEN, "test_shm_mq");
226 : 5 : worker.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
227 : : /* set bgw_notify_pid, so we can detect if the worker stops */
228 : 5 : worker.bgw_notify_pid = MyProcPid;
229 : :
230 : : /* Register the workers. */
231 [ + + ]: 12 : for (i = 0; i < nworkers; ++i)
232 : : {
233 : 7 : snprintf(worker.bgw_name, BGW_MAXLEN, "test_shm_mq worker %d", i + 1);
234 [ - + ]: 7 : if (!RegisterDynamicBackgroundWorker(&worker, &wstate->handle[i]))
235 [ # # ]: 0 : ereport(ERROR,
236 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
237 : : errmsg("could not register background process"),
238 : : errhint("You may need to increase \"max_worker_processes\".")));
239 : 7 : ++wstate->nworkers;
240 : : }
241 : :
242 : : /* All done. */
243 : 5 : MemoryContextSwitchTo(oldcontext);
244 : 5 : return wstate;
245 : : }
246 : :
247 : : static void
248 : 0 : cleanup_background_workers(dsm_segment *seg, Datum arg)
249 : : {
250 : 0 : worker_state *wstate = (worker_state *) DatumGetPointer(arg);
251 : :
252 [ # # ]: 0 : while (wstate->nworkers > 0)
253 : : {
254 : 0 : --wstate->nworkers;
255 : 0 : TerminateBackgroundWorker(wstate->handle[wstate->nworkers]);
256 : : }
257 : 0 : }
258 : :
259 : : static void
260 : 5 : wait_for_workers_to_become_ready(worker_state *wstate,
261 : : volatile test_shm_mq_header *hdr)
262 : : {
263 : 5 : bool result = false;
264 : :
265 : : for (;;)
266 : 15 : {
267 : : int workers_ready;
268 : :
269 : : /* If all the workers are ready, we have succeeded. */
270 : 20 : SpinLockAcquire(&hdr->mutex);
271 : 20 : workers_ready = hdr->workers_ready;
272 : 20 : SpinLockRelease(&hdr->mutex);
273 [ + + ]: 20 : if (workers_ready >= wstate->nworkers)
274 : : {
275 : 5 : result = true;
276 : 5 : break;
277 : : }
278 : :
279 : : /* If any workers (or the postmaster) have died, we have failed. */
280 [ - + ]: 15 : if (!check_worker_status(wstate))
281 : : {
282 : 0 : result = false;
283 : 0 : break;
284 : : }
285 : :
286 : : /* first time, allocate or get the custom wait event */
287 [ + + ]: 15 : if (we_bgworker_startup == 0)
288 : 1 : we_bgworker_startup = WaitEventExtensionNew("TestShmMqBgWorkerStartup");
289 : :
290 : : /* Wait to be signaled. */
291 : 15 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
292 : : we_bgworker_startup);
293 : :
294 : : /* Reset the latch so we don't spin. */
295 : 15 : ResetLatch(MyLatch);
296 : :
297 : : /* An interrupt may have occurred while we were waiting. */
298 [ - + ]: 15 : CHECK_FOR_INTERRUPTS();
299 : : }
300 : :
301 [ - + ]: 5 : if (!result)
302 [ # # ]: 0 : ereport(ERROR,
303 : : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
304 : : errmsg("one or more background workers failed to start")));
305 : 5 : }
306 : :
307 : : static bool
308 : 15 : check_worker_status(worker_state *wstate)
309 : : {
310 : : int n;
311 : :
312 : : /* If any workers (or the postmaster) have died, we have failed. */
313 [ + + ]: 36 : for (n = 0; n < wstate->nworkers; ++n)
314 : : {
315 : : BgwHandleStatus status;
316 : : pid_t pid;
317 : :
318 : 21 : status = GetBackgroundWorkerPid(wstate->handle[n], &pid);
319 [ + - - + ]: 21 : if (status == BGWH_STOPPED || status == BGWH_POSTMASTER_DIED)
320 : 0 : return false;
321 : : }
322 : :
323 : : /* Otherwise, things still look OK. */
324 : 15 : return true;
325 : : }
|
