In SQLITE_ENABLE_SETLK_TIMEOUT builds, use blocking locks in place of sleep() when opening a read-transaction.

FossilOrigin-Name: a51ef39998e25e86bd0600e71d15011b12e05f4319608018293bdaecb09e8c97

manifest

@@ -1,5 +1,5 @@
-C Add\sthe\s--buildonly\sand\s--dryrun\soptions\sto\stestrunner.tcl.
+C In\sSQLITE_ENABLE_SETLK_TIMEOUT\sbuilds,\suse\sblocking\slocks\sin\splace\sof\ssleep()\swhen\sopening\sa\sread-transaction.
-D 2023-11-24T20:14:18.100
+D 2023-11-27T19:22:50.016
 F .fossil-settings/empty-dirs dbb81e8fc0401ac46a1491ab34a7f2c7c0452f2f06b54ebb845d024ca8283ef1
 F .fossil-settings/ignore-glob 35175cdfcf539b2318cb04a9901442804be81cd677d8b889fcc9149c21f239ea
 F LICENSE.md df5091916dbb40e6e9686186587125e1b2ff51f022cc334e886c19a0e9982724
@@ -712,7 +712,7 @@ F src/os.h 1ff5ae51d339d0e30d8a9d814f4b8f8e448169304d83a7ed9db66a65732f3e63
 F src/os_common.h 6c0eb8dd40ef3e12fe585a13e709710267a258e2c8dd1c40b1948a1d14582e06
 F src/os_kv.c 4d39e1f1c180b11162c6dc4aa8ad34053873a639bac6baae23272fc03349986a
 F src/os_setup.h 6011ad7af5db4e05155f385eb3a9b4470688de6f65d6166b8956e58a3d872107
-F src/os_unix.c dc5404b56da7fb13cf272ddb94c3753cf9e82d32a65cba35dbb6aadcb849419c
+F src/os_unix.c 97bdcd43315da7aaec9fea2da1ff7c9de458f93dd363e073f2742403a7f2e011
 F src/os_win.c 4a50a154aeebc66a1f8fb79c1ff6dd5fe3d005556533361e0d460d41cb6a45a8
 F src/os_win.h 7b073010f1451abe501be30d12f6bc599824944a
 F src/pager.c 987ab3a2cd9065d62e9955474470ff733445e2357432a67e3d0f5a8f9313e334
@@ -810,7 +810,7 @@ F src/vdbetrace.c fe0bc29ebd4e02c8bc5c1945f1d2e6be5927ec12c06d89b03ef2a4def34bf8
 F src/vdbevtab.c 2143db7db0ceed69b21422581f434baffc507a08d831565193a7a02882a1b6a7
 F src/vtab.c 154725ebecd3bc02f7fbd7ad3974334f73fff76e02a964e828e48a7c5fb7efff
 F src/vxworks.h d2988f4e5a61a4dfe82c6524dd3d6e4f2ce3cdb9
-F src/wal.c 69aa3ce60b2862a24cd86bb528e653e2137388ead258ef64db49ec9038807f5f
+F src/wal.c f64bf6799dbbd346a9e2aae9b96e51dbbb222d24a7e7509db85fb0144e8d9f38
 F src/wal.h ba252daaa94f889f4b2c17c027e823d9be47ce39da1d3799886bbd51f0490452
 F src/walker.c 7c7ea0115345851c3da4e04e2e239a29983b61fb5b038b94eede6aba462640e2
 F src/where.c 1fdc69ce1333e9bd6d7d3df9fa5af1373a3f5bfdd52108d1dbc0ca85a55f777e
@@ -2143,8 +2143,11 @@ F vsixtest/vsixtest.tcl 6a9a6ab600c25a91a7acc6293828957a386a8a93
 F vsixtest/vsixtest.vcxproj.data 2ed517e100c66dc455b492e1a33350c1b20fbcdc
 F vsixtest/vsixtest.vcxproj.filters 37e51ffedcdb064aad6ff33b6148725226cd608e
 F vsixtest/vsixtest_TemporaryKey.pfx e5b1b036facdb453873e7084e1cae9102ccc67a0
-P f8ea0b58b37f4052ded448e595d6d2992988a33b8ecfe58d68f20532f8cb5a37
+P a0c87ae9d3db914d18e2c8811db0d0ae3ad7b15c63de84fa975efce28bace27e
-R a09b448e8fa45e3193b13a1fe5b499b2
+R 35e10752e16b85b0bde20be6e18f8d14
+T *branch * blocking-dms-lock
+T *sym-blocking-dms-lock *
+T -sym-trunk *
 U dan
-Z ce0098ca9d7841ef7ea900cfd937f5a5
+Z fd8676cc431af47934a9227125e3da50
 # Remove this line to create a well-formed Fossil manifest.

manifest.uuid

@@ -1 +1 @@
-a0c87ae9d3db914d18e2c8811db0d0ae3ad7b15c63de84fa975efce28bace27e
+a51ef39998e25e86bd0600e71d15011b12e05f4319608018293bdaecb09e8c97

src/os_unix.c

@@ -4595,7 +4595,20 @@ static int unixLockSharedMemory(unixFile *pDbFd, unixShmNode *pShmNode){
       pShmNode->isUnlocked = 1;
       rc = SQLITE_READONLY_CANTINIT;
     }else{
+#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+      /* Do not use a blocking lock here. If the lock cannot be obtained
+      ** immediately, it means some other connection is truncating the
+      ** *-shm file. And after it has done so, it will not release its
+      ** lock, but only downgrade it to a shared lock. So no point in
+      ** blocking here. The call below to obtain the shared DMS lock may 
+      ** use a blocking lock. */
+      int iSaveTimeout = pDbFd->iBusyTimeout;
+      pDbFd->iBusyTimeout = 0;
+#endif
       rc = unixShmSystemLock(pDbFd, F_WRLCK, UNIX_SHM_DMS, 1);
+#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+      pDbFd->iBusyTimeout = iSaveTimeout;
+#endif
       /* The first connection to attach must truncate the -shm file.  We
       ** truncate to 3 bytes (an arbitrary small number, less than the
       ** -shm header size) rather than 0 as a system debugging aid, to

src/wal.c

@@ -2004,6 +2004,19 @@ static int walIteratorInit(Wal *pWal, u32 nBackfill, WalIterator **pp){
 }
 
 #ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+
+
+/*
+** Attempt to enable blocking locks that block for nMs ms. Return 1 if 
+** blocking locks are successfully enabled, or 0 otherwise.
+*/
+static int walEnableBlockingMs(Wal *pWal, int nMs){
+  int rc = sqlite3OsFileControl(
+      pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&nMs
+  );
+  return (rc==SQLITE_OK);
+}
+
 /*
 ** Attempt to enable blocking locks. Blocking locks are enabled only if (a)
 ** they are supported by the VFS, and (b) the database handle is configured
@@ -2015,11 +2028,7 @@ static int walEnableBlocking(Wal *pWal){
   if( pWal->db ){
     int tmout = pWal->db->busyTimeout;
     if( tmout ){
-      int rc;
+      res = walEnableBlockingMs(pWal, tmout);
-      rc = sqlite3OsFileControl(
-          pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout
-      );
-      res = (rc==SQLITE_OK);
     }
   }
   return res;
@@ -2068,20 +2077,10 @@ void sqlite3WalDb(Wal *pWal, sqlite3 *db){
   pWal->db = db;
 }
 
-/*
-** Take an exclusive WRITE lock. Blocking if so configured.
-*/
-static int walLockWriter(Wal *pWal){
-  int rc;
-  walEnableBlocking(pWal);
-  rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1);
-  walDisableBlocking(pWal);
-  return rc;
-}
 #else
 # define walEnableBlocking(x) 0
 # define walDisableBlocking(x)
-# define walLockWriter(pWal) walLockExclusive((pWal), WAL_WRITE_LOCK, 1)
+# define walEnableBlockingMs(pWal, ms) 0
 # define sqlite3WalDb(pWal, db)
 #endif   /* ifdef SQLITE_ENABLE_SETLK_TIMEOUT */
 
@@ -2682,7 +2681,9 @@ static int walIndexReadHdr(Wal *pWal, int *pChanged){
       }
     }else{
       int bWriteLock = pWal->writeLock;
-      if( bWriteLock || SQLITE_OK==(rc = walLockWriter(pWal)) ){
+      if( bWriteLock 
+       || SQLITE_OK==(rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1)) 
+      ){
         pWal->writeLock = 1;
         if( SQLITE_OK==(rc = walIndexPage(pWal, 0, &page0)) ){
           badHdr = walIndexTryHdr(pWal, pChanged);
@@ -2690,7 +2691,8 @@ static int walIndexReadHdr(Wal *pWal, int *pChanged){
             /* If the wal-index header is still malformed even while holding
             ** a WRITE lock, it can only mean that the header is corrupted and
             ** needs to be reconstructed.  So run recovery to do exactly that.
-            */
+            ** Disable blocking locks first.  */
+            walDisableBlocking(pWal);
             rc = walIndexRecover(pWal);
             *pChanged = 1;
           }
@@ -2987,6 +2989,17 @@ static int walTryBeginRead(Wal *pWal, int *pChanged, int useWal, int cnt){
       return SQLITE_PROTOCOL;
     }
     if( cnt>=10 ) nDelay = (cnt-9)*(cnt-9)*39;
+#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+    /* In SQLITE_ENABLE_SETLK_TIMEOUT builds, configure the file-descriptor
+    ** to block for locks for approximately nDelay us. This affects two
+    ** locks (a) the WRITER lock taken in walIndexReadHdr() if the first
+    ** attempted read fails and (b) the shared lock taken on the DMS slot
+    ** in os_unix.c.  Both of these locks are attempted from within the 
+    ** call to walIndexReadHdr() below.  */
+    if( cnt>=10 && !useWal && walEnableBlockingMs(pWal, (nDelay+999)/1000) ){
+      nDelay = 1;
+    }
+#endif
     sqlite3OsSleep(pWal->pVfs, nDelay);
   }
 
@@ -2995,6 +3008,10 @@ static int walTryBeginRead(Wal *pWal, int *pChanged, int useWal, int cnt){
     if( pWal->bShmUnreliable==0 ){
       rc = walIndexReadHdr(pWal, pChanged);
     }
+#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+    walDisableBlocking(pWal);
+    if( rc==SQLITE_BUSY_TIMEOUT ) rc = SQLITE_BUSY;
+#endif
     if( rc==SQLITE_BUSY ){
       /* If there is not a recovery running in another thread or process
       ** then convert BUSY errors to WAL_RETRY.  If recovery is known to