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) 2013-2023, 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/procsignal.h"
22 : #include "storage/shm_toc.h"
23 : #include "test_shm_mq.h"
24 : #include "utils/memutils.h"
25 :
26 : typedef struct
27 : {
28 : int nworkers;
29 : BackgroundWorkerHandle *handle[FLEXIBLE_ARRAY_MEMBER];
30 : } worker_state;
31 :
32 : static void setup_dynamic_shared_memory(int64 queue_size, int nworkers,
33 : dsm_segment **segp,
34 : test_shm_mq_header **hdrp,
35 : shm_mq **outp, shm_mq **inp);
36 : static worker_state *setup_background_workers(int nworkers,
37 : dsm_segment *seg);
38 : static void cleanup_background_workers(dsm_segment *seg, Datum arg);
39 : static void wait_for_workers_to_become_ready(worker_state *wstate,
40 : volatile test_shm_mq_header *hdr);
41 : static bool check_worker_status(worker_state *wstate);
42 :
43 : /* value cached, fetched from shared memory */
44 : static uint32 we_bgworker_startup = 0;
45 :
46 : /*
47 : * Set up a dynamic shared memory segment and zero or more background workers
48 : * for a test run.
49 : */
50 : void
51 10 : test_shm_mq_setup(int64 queue_size, int32 nworkers, dsm_segment **segp,
52 : shm_mq_handle **output, shm_mq_handle **input)
53 : {
54 : dsm_segment *seg;
55 : test_shm_mq_header *hdr;
56 10 : shm_mq *outq = NULL; /* placate compiler */
57 10 : shm_mq *inq = NULL; /* placate compiler */
58 : worker_state *wstate;
59 :
60 : /* Set up a dynamic shared memory segment. */
61 10 : setup_dynamic_shared_memory(queue_size, nworkers, &seg, &hdr, &outq, &inq);
62 10 : *segp = seg;
63 :
64 : /* Register background workers. */
65 10 : wstate = setup_background_workers(nworkers, seg);
66 :
67 : /* Attach the queues. */
68 10 : *output = shm_mq_attach(outq, seg, wstate->handle[0]);
69 10 : *input = shm_mq_attach(inq, seg, wstate->handle[nworkers - 1]);
70 :
71 : /* Wait for workers to become ready. */
72 10 : wait_for_workers_to_become_ready(wstate, hdr);
73 :
74 : /*
75 : * Once we reach this point, all workers are ready. We no longer need to
76 : * kill them if we die; they'll die on their own as the message queues
77 : * shut down.
78 : */
79 10 : cancel_on_dsm_detach(seg, cleanup_background_workers,
80 : PointerGetDatum(wstate));
81 10 : pfree(wstate);
82 10 : }
83 :
84 : /*
85 : * Set up a dynamic shared memory segment.
86 : *
87 : * We set up a small control region that contains only a test_shm_mq_header,
88 : * plus one region per message queue. There are as many message queues as
89 : * the number of workers, plus one.
90 : */
91 : static void
92 10 : setup_dynamic_shared_memory(int64 queue_size, int nworkers,
93 : dsm_segment **segp, test_shm_mq_header **hdrp,
94 : shm_mq **outp, shm_mq **inp)
95 : {
96 : shm_toc_estimator e;
97 : int i;
98 : Size segsize;
99 : dsm_segment *seg;
100 : shm_toc *toc;
101 : test_shm_mq_header *hdr;
102 :
103 : /* Ensure a valid queue size. */
104 10 : if (queue_size < 0 || ((uint64) queue_size) < shm_mq_minimum_size)
105 0 : ereport(ERROR,
106 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
107 : errmsg("queue size must be at least %zu bytes",
108 : shm_mq_minimum_size)));
109 : if (queue_size != ((Size) queue_size))
110 : ereport(ERROR,
111 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
112 : errmsg("queue size overflows size_t")));
113 :
114 : /*
115 : * Estimate how much shared memory we need.
116 : *
117 : * Because the TOC machinery may choose to insert padding of oddly-sized
118 : * requests, we must estimate each chunk separately.
119 : *
120 : * We need one key to register the location of the header, and we need
121 : * nworkers + 1 keys to track the locations of the message queues.
122 : */
123 10 : shm_toc_initialize_estimator(&e);
124 10 : shm_toc_estimate_chunk(&e, sizeof(test_shm_mq_header));
125 34 : for (i = 0; i <= nworkers; ++i)
126 24 : shm_toc_estimate_chunk(&e, (Size) queue_size);
127 10 : shm_toc_estimate_keys(&e, 2 + nworkers);
128 10 : segsize = shm_toc_estimate(&e);
129 :
130 : /* Create the shared memory segment and establish a table of contents. */
131 10 : seg = dsm_create(shm_toc_estimate(&e), 0);
132 10 : toc = shm_toc_create(PG_TEST_SHM_MQ_MAGIC, dsm_segment_address(seg),
133 : segsize);
134 :
135 : /* Set up the header region. */
136 10 : hdr = shm_toc_allocate(toc, sizeof(test_shm_mq_header));
137 10 : SpinLockInit(&hdr->mutex);
138 10 : hdr->workers_total = nworkers;
139 10 : hdr->workers_attached = 0;
140 10 : hdr->workers_ready = 0;
141 10 : shm_toc_insert(toc, 0, hdr);
142 :
143 : /* Set up one message queue per worker, plus one. */
144 34 : for (i = 0; i <= nworkers; ++i)
145 : {
146 : shm_mq *mq;
147 :
148 24 : mq = shm_mq_create(shm_toc_allocate(toc, (Size) queue_size),
149 : (Size) queue_size);
150 24 : shm_toc_insert(toc, i + 1, mq);
151 :
152 24 : if (i == 0)
153 : {
154 : /* We send messages to the first queue. */
155 10 : shm_mq_set_sender(mq, MyProc);
156 10 : *outp = mq;
157 : }
158 24 : if (i == nworkers)
159 : {
160 : /* We receive messages from the last queue. */
161 10 : shm_mq_set_receiver(mq, MyProc);
162 10 : *inp = mq;
163 : }
164 : }
165 :
166 : /* Return results to caller. */
167 10 : *segp = seg;
168 10 : *hdrp = hdr;
169 10 : }
170 :
171 : /*
172 : * Register background workers.
173 : */
174 : static worker_state *
175 10 : setup_background_workers(int nworkers, dsm_segment *seg)
176 : {
177 : MemoryContext oldcontext;
178 : BackgroundWorker worker;
179 : worker_state *wstate;
180 : int i;
181 :
182 : /*
183 : * We need the worker_state object and the background worker handles to
184 : * which it points to be allocated in CurTransactionContext rather than
185 : * ExprContext; otherwise, they'll be destroyed before the on_dsm_detach
186 : * hooks run.
187 : */
188 10 : oldcontext = MemoryContextSwitchTo(CurTransactionContext);
189 :
190 : /* Create worker state object. */
191 10 : wstate = MemoryContextAlloc(TopTransactionContext,
192 10 : offsetof(worker_state, handle) +
193 : sizeof(BackgroundWorkerHandle *) * nworkers);
194 10 : wstate->nworkers = 0;
195 :
196 : /*
197 : * Arrange to kill all the workers if we abort before all workers are
198 : * finished hooking themselves up to the dynamic shared memory segment.
199 : *
200 : * If we die after all the workers have finished hooking themselves up to
201 : * the dynamic shared memory segment, we'll mark the two queues to which
202 : * we're directly connected as detached, and the worker(s) connected to
203 : * those queues will exit, marking any other queues to which they are
204 : * connected as detached. This will cause any as-yet-unaware workers
205 : * connected to those queues to exit in their turn, and so on, until
206 : * everybody exits.
207 : *
208 : * But suppose the workers which are supposed to connect to the queues to
209 : * which we're directly attached exit due to some error before they
210 : * actually attach the queues. The remaining workers will have no way of
211 : * knowing this. From their perspective, they're still waiting for those
212 : * workers to start, when in fact they've already died.
213 : */
214 10 : on_dsm_detach(seg, cleanup_background_workers,
215 : PointerGetDatum(wstate));
216 :
217 : /* Configure a worker. */
218 10 : memset(&worker, 0, sizeof(worker));
219 10 : worker.bgw_flags = BGWORKER_SHMEM_ACCESS;
220 10 : worker.bgw_start_time = BgWorkerStart_ConsistentState;
221 10 : worker.bgw_restart_time = BGW_NEVER_RESTART;
222 10 : sprintf(worker.bgw_library_name, "test_shm_mq");
223 10 : sprintf(worker.bgw_function_name, "test_shm_mq_main");
224 10 : snprintf(worker.bgw_type, BGW_MAXLEN, "test_shm_mq");
225 10 : worker.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
226 : /* set bgw_notify_pid, so we can detect if the worker stops */
227 10 : worker.bgw_notify_pid = MyProcPid;
228 :
229 : /* Register the workers. */
230 24 : for (i = 0; i < nworkers; ++i)
231 : {
232 14 : if (!RegisterDynamicBackgroundWorker(&worker, &wstate->handle[i]))
233 0 : ereport(ERROR,
234 : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
235 : errmsg("could not register background process"),
236 : errhint("You may need to increase max_worker_processes.")));
237 14 : ++wstate->nworkers;
238 : }
239 :
240 : /* All done. */
241 10 : MemoryContextSwitchTo(oldcontext);
242 10 : return wstate;
243 : }
244 :
245 : static void
246 0 : cleanup_background_workers(dsm_segment *seg, Datum arg)
247 : {
248 0 : worker_state *wstate = (worker_state *) DatumGetPointer(arg);
249 :
250 0 : while (wstate->nworkers > 0)
251 : {
252 0 : --wstate->nworkers;
253 0 : TerminateBackgroundWorker(wstate->handle[wstate->nworkers]);
254 : }
255 0 : }
256 :
257 : static void
258 10 : wait_for_workers_to_become_ready(worker_state *wstate,
259 : volatile test_shm_mq_header *hdr)
260 : {
261 10 : bool result = false;
262 :
263 : for (;;)
264 54 : {
265 : int workers_ready;
266 :
267 : /* If all the workers are ready, we have succeeded. */
268 64 : SpinLockAcquire(&hdr->mutex);
269 64 : workers_ready = hdr->workers_ready;
270 64 : SpinLockRelease(&hdr->mutex);
271 64 : if (workers_ready >= wstate->nworkers)
272 : {
273 10 : result = true;
274 10 : break;
275 : }
276 :
277 : /* If any workers (or the postmaster) have died, we have failed. */
278 54 : if (!check_worker_status(wstate))
279 : {
280 0 : result = false;
281 0 : break;
282 : }
283 :
284 : /* first time, allocate or get the custom wait event */
285 54 : if (we_bgworker_startup == 0)
286 2 : we_bgworker_startup = WaitEventExtensionNew("TestShmMqBgWorkerStartup");
287 :
288 : /* Wait to be signaled. */
289 54 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
290 : we_bgworker_startup);
291 :
292 : /* Reset the latch so we don't spin. */
293 54 : ResetLatch(MyLatch);
294 :
295 : /* An interrupt may have occurred while we were waiting. */
296 54 : CHECK_FOR_INTERRUPTS();
297 : }
298 :
299 10 : if (!result)
300 0 : ereport(ERROR,
301 : (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
302 : errmsg("one or more background workers failed to start")));
303 10 : }
304 :
305 : static bool
306 54 : check_worker_status(worker_state *wstate)
307 : {
308 : int n;
309 :
310 : /* If any workers (or the postmaster) have died, we have failed. */
311 140 : for (n = 0; n < wstate->nworkers; ++n)
312 : {
313 : BgwHandleStatus status;
314 : pid_t pid;
315 :
316 86 : status = GetBackgroundWorkerPid(wstate->handle[n], &pid);
317 86 : if (status == BGWH_STOPPED || status == BGWH_POSTMASTER_DIED)
318 0 : return false;
319 : }
320 :
321 : /* Otherwise, things still look OK. */
322 54 : return true;
323 : }
|
