src/backend/postmaster/walwriter.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * walwriter.c
   4  *
   5  * The WAL writer background process is new as of Postgres 8.3.  It attempts
   6  * to keep regular backends from having to write out (and fsync) WAL pages.
   7  * Also, it guarantees that transaction commit records that weren't synced
   8  * to disk immediately upon commit (ie, were "asynchronously committed")
   9  * will reach disk within a knowable time --- which, as it happens, is at
  10  * most three times the wal_writer_delay cycle time.
  11  *
  12  * Note that as with the bgwriter for shared buffers, regular backends are
  13  * still empowered to issue WAL writes and fsyncs when the walwriter doesn't
  14  * keep up. This means that the WALWriter is not an essential process and
  15  * can shutdown quickly when requested.
  16  *
  17  * Because the walwriter's cycle is directly linked to the maximum delay
  18  * before async-commit transactions are guaranteed committed, it's probably
  19  * unwise to load additional functionality onto it.  For instance, if you've
  20  * got a yen to create xlog segments further in advance, that'd be better done
  21  * in bgwriter than in walwriter.
  22  *
  23  * The walwriter is started by the postmaster as soon as the startup subprocess
  24  * finishes.  It remains alive until the postmaster commands it to terminate.
  25  * Normal termination is by SIGTERM, which instructs the walwriter to exit(0).
  26  * Emergency termination is by SIGQUIT; like any backend, the walwriter will
  27  * simply abort and exit on SIGQUIT.
  28  *
  29  * If the walwriter exits unexpectedly, the postmaster treats that the same
  30  * as a backend crash: shared memory may be corrupted, so remaining backends
  31  * should be killed by SIGQUIT and then a recovery cycle started.
  32  *
  33  *
  34  * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
  35  *
  36  *
  37  * IDENTIFICATION
  38  *        src/backend/postmaster/walwriter.c
  39  *
  40  *-------------------------------------------------------------------------
  41  */
  42 #include "postgres.h"
  43
  44 #include <signal.h>
  45 #include <unistd.h>
  46
  47 #include "access/xlog.h"
  48 #include "libpq/pqsignal.h"
  49 #include "miscadmin.h"
  50 #include "pgstat.h"
  51 #include "postmaster/interrupt.h"
  52 #include "postmaster/walwriter.h"
  53 #include "storage/bufmgr.h"
  54 #include "storage/condition_variable.h"
  55 #include "storage/fd.h"
  56 #include "storage/ipc.h"
  57 #include "storage/lwlock.h"
  58 #include "storage/proc.h"
  59 #include "storage/procsignal.h"
  60 #include "storage/smgr.h"
  61 #include "utils/guc.h"
  62 #include "utils/hsearch.h"
  63 #include "utils/memutils.h"
  64 #include "utils/resowner.h"
  65
  66
  67 /*
  68  * GUC parameters
  69  */
  70 int                     WalWriterDelay = 200;
  71 int                     WalWriterFlushAfter = 128;
  72
  73 /*
  74  * Number of do-nothing loops before lengthening the delay time, and the
  75  * multiplier to apply to WalWriterDelay when we do decide to hibernate.
  76  * (Perhaps these need to be configurable?)
  77  */
  78 #define LOOPS_UNTIL_HIBERNATE           50
  79 #define HIBERNATE_FACTOR                        25
  80
  81 /* Prototypes for private functions */
  82 static void HandleWalWriterInterrupts(void);
  83
  84 /*
  85  * Main entry point for walwriter process
  86  *
  87  * This is invoked from AuxiliaryProcessMain, which has already created the
  88  * basic execution environment, but not enabled signals yet.
  89  */
  90 void
  91 WalWriterMain(void)
  92 {
  93         sigjmp_buf      local_sigjmp_buf;
  94         MemoryContext walwriter_context;
  95         int                     left_till_hibernate;
  96         bool            hibernating;
  97
  98         /*
  99          * Properly accept or ignore signals the postmaster might send us
 100          *
 101          * We have no particular use for SIGINT at the moment, but seems
 102          * reasonable to treat like SIGTERM.
 103          */
 104         pqsignal(SIGHUP, SignalHandlerForConfigReload);
 105         pqsignal(SIGINT, SignalHandlerForShutdownRequest);
 106         pqsignal(SIGTERM, SignalHandlerForShutdownRequest);
 107         /* SIGQUIT handler was already set up by InitPostmasterChild */
 108         pqsignal(SIGALRM, SIG_IGN);
 109         pqsignal(SIGPIPE, SIG_IGN);
 110         pqsignal(SIGUSR1, procsignal_sigusr1_handler);
 111         pqsignal(SIGUSR2, SIG_IGN); /* not used */
 112
 113         /*
 114          * Reset some signals that are accepted by postmaster but not here
 115          */
 116         pqsignal(SIGCHLD, SIG_DFL);
 117
 118         /*
 119          * Create a memory context that we will do all our work in.  We do this so
 120          * that we can reset the context during error recovery and thereby avoid
 121          * possible memory leaks.  Formerly this code just ran in
 122          * TopMemoryContext, but resetting that would be a really bad idea.
 123          */
 124         walwriter_context = AllocSetContextCreate(TopMemoryContext,
 125                                                                                           "Wal Writer",
 126                                                                                           ALLOCSET_DEFAULT_SIZES);
 127         MemoryContextSwitchTo(walwriter_context);
 128
 129         /*
 130          * If an exception is encountered, processing resumes here.
 131          *
 132          * You might wonder why this isn't coded as an infinite loop around a
 133          * PG_TRY construct.  The reason is that this is the bottom of the
 134          * exception stack, and so with PG_TRY there would be no exception handler
 135          * in force at all during the CATCH part.  By leaving the outermost setjmp
 136          * always active, we have at least some chance of recovering from an error
 137          * during error recovery.  (If we get into an infinite loop thereby, it
 138          * will soon be stopped by overflow of elog.c's internal state stack.)
 139          *
 140          * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
 141          * (to wit, BlockSig) will be restored when longjmp'ing to here.  Thus,
 142          * signals other than SIGQUIT will be blocked until we complete error
 143          * recovery.  It might seem that this policy makes the HOLD_INTERRUPTS()
 144          * call redundant, but it is not since InterruptPending might be set
 145          * already.
 146          */
 147         if (sigsetjmp(local_sigjmp_buf, 1) != 0)
 148         {
 149                 /* Since not using PG_TRY, must reset error stack by hand */
 150                 error_context_stack = NULL;
 151
 152                 /* Prevent interrupts while cleaning up */
 153                 HOLD_INTERRUPTS();
 154
 155                 /* Report the error to the server log */
 156                 EmitErrorReport();
 157
 158                 /*
 159                  * These operations are really just a minimal subset of
 160                  * AbortTransaction().  We don't have very many resources to worry
 161                  * about in walwriter, but we do have LWLocks, and perhaps buffers?
 162                  */
 163                 LWLockReleaseAll();
 164                 ConditionVariableCancelSleep();
 165                 pgstat_report_wait_end();
 166                 AbortBufferIO();
 167                 UnlockBuffers();
 168                 ReleaseAuxProcessResources(false);
 169                 AtEOXact_Buffers(false);
 170                 AtEOXact_SMgr();
 171                 AtEOXact_Files(false);
 172                 AtEOXact_HashTables(false);
 173
 174                 /*
 175                  * Now return to normal top-level context and clear ErrorContext for
 176                  * next time.
 177                  */
 178                 MemoryContextSwitchTo(walwriter_context);
 179                 FlushErrorState();
 180
 181                 /* Flush any leaked data in the top-level context */
 182                 MemoryContextResetAndDeleteChildren(walwriter_context);
 183
 184                 /* Now we can allow interrupts again */
 185                 RESUME_INTERRUPTS();
 186
 187                 /*
 188                  * Sleep at least 1 second after any error.  A write error is likely
 189                  * to be repeated, and we don't want to be filling the error logs as
 190                  * fast as we can.
 191                  */
 192                 pg_usleep(1000000L);
 193
 194                 /*
 195                  * Close all open files after any error.  This is helpful on Windows,
 196                  * where holding deleted files open causes various strange errors.
 197                  * It's not clear we need it elsewhere, but shouldn't hurt.
 198                  */
 199                 smgrcloseall();
 200         }
 201
 202         /* We can now handle ereport(ERROR) */
 203         PG_exception_stack = &local_sigjmp_buf;
 204
 205         /*
 206          * Unblock signals (they were blocked when the postmaster forked us)
 207          */
 208         PG_SETMASK(&UnBlockSig);
 209
 210         /*
 211          * Reset hibernation state after any error.
 212          */
 213         left_till_hibernate = LOOPS_UNTIL_HIBERNATE;
 214         hibernating = false;
 215         SetWalWriterSleeping(false);
 216
 217         /*
 218          * Advertise our latch that backends can use to wake us up while we're
 219          * sleeping.
 220          */
 221         ProcGlobal->walwriterLatch = &MyProc->procLatch;
 222
 223         /*
 224          * Loop forever
 225          */
 226         for (;;)
 227         {
 228                 long            cur_timeout;
 229
 230                 /*
 231                  * Advertise whether we might hibernate in this cycle.  We do this
 232                  * before resetting the latch to ensure that any async commits will
 233                  * see the flag set if they might possibly need to wake us up, and
 234                  * that we won't miss any signal they send us.  (If we discover work
 235                  * to do in the last cycle before we would hibernate, the global flag
 236                  * will be set unnecessarily, but little harm is done.)  But avoid
 237                  * touching the global flag if it doesn't need to change.
 238                  */
 239                 if (hibernating != (left_till_hibernate <= 1))
 240                 {
 241                         hibernating = (left_till_hibernate <= 1);
 242                         SetWalWriterSleeping(hibernating);
 243                 }
 244
 245                 /* Clear any already-pending wakeups */
 246                 ResetLatch(MyLatch);
 247
 248                 /* Process any signals received recently */
 249                 HandleWalWriterInterrupts();
 250
 251                 /*
 252                  * Do what we're here for; then, if XLogBackgroundFlush() found useful
 253                  * work to do, reset hibernation counter.
 254                  */
 255                 if (XLogBackgroundFlush())
 256                         left_till_hibernate = LOOPS_UNTIL_HIBERNATE;
 257                 else if (left_till_hibernate > 0)
 258                         left_till_hibernate--;
 259
 260                 /* Send WAL statistics to the stats collector */
 261                 pgstat_send_wal(false);
 262
 263                 /*
 264                  * Sleep until we are signaled or WalWriterDelay has elapsed.  If we
 265                  * haven't done anything useful for quite some time, lengthen the
 266                  * sleep time so as to reduce the server's idle power consumption.
 267                  */
 268                 if (left_till_hibernate > 0)
 269                         cur_timeout = WalWriterDelay;   /* in ms */
 270                 else
 271                         cur_timeout = WalWriterDelay * HIBERNATE_FACTOR;
 272
 273                 (void) WaitLatch(MyLatch,
 274                                                  WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
 275                                                  cur_timeout,
 276                                                  WAIT_EVENT_WAL_WRITER_MAIN);
 277         }
 278 }
 279
 280 /*
 281  * Interrupt handler for main loops of WAL writer process.
 282  */
 283 static void
 284 HandleWalWriterInterrupts(void)
 285 {
 286         if (ProcSignalBarrierPending)
 287                 ProcessProcSignalBarrier();
 288
 289         if (ConfigReloadPending)
 290         {
 291                 ConfigReloadPending = false;
 292                 ProcessConfigFile(PGC_SIGHUP);
 293         }
 294
 295         if (ShutdownRequestPending)
 296         {
 297                 /*
 298                  * Force to send remaining WAL statistics to the stats collector at
 299                  * process exit.
 300                  *
 301                  * Since pgstat_send_wal is invoked with 'force' is false in main loop
 302                  * to avoid overloading to the stats collector, there may exist unsent
 303                  * stats counters for the WAL writer.
 304                  */
 305                 pgstat_send_wal(true);
 306
 307                 proc_exit(0);
 308         }
 309 }