ubik sync client error recovery

[openafs.git] / src / ubik / recovery.c

diff --git a/src/ubik/recovery.c b/src/ubik/recovery.c
index e23d31c..cc5d951 100644 (file)
--- a/src/ubik/recovery.c
+++ b/src/ubik/recovery.c
@@ -10,6 +10,7 @@
 #include <afsconfig.h>
 #include <afs/param.h>
 
+#include <roken.h>
 
 #include <sys/types.h>
 #include <string.h>
@@ -30,6 +31,7 @@
 #include <rx/xdr.h>
 #include <rx/rx.h>
 #include <afs/afsutil.h>
+#include <afs/cellconfig.h>
 
 #define UBIK_INTERNALS
 #include "ubik.h"
@@ -83,13 +85,15 @@ urecovery_ResetState(void)
  * \brief sync site
  *
  * routine called when a non-sync site server goes down; restarts recovery
- * process to send missing server the new db when it comes back up.
+ * process to send missing server the new db when it comes back up for
+ * non-sync site servers.
  *
  * \note This routine should not do anything with variables used by non-sync site servers.
  */
 int
-urecovery_LostServer(void)
+urecovery_LostServer(struct ubik_server *ts)
 {
+    ubeacon_ReinitServer(ts);
 #if !defined(AFS_PTHREAD_ENV)
     /*  No corresponding LWP_WaitProcess found anywhere for this -- klm */
     LWP_NoYieldSignal(&urecovery_state);
@@ -276,15 +280,21 @@ ReplayLog(struct ubik_dbase *adbase)
            } else if (opcode == LOGABORT)
                panic("log abort\n");
            else if (opcode == LOGEND) {
+               struct ubik_version version;
                tpos += 4;
                code =
                    (*adbase->read) (adbase, LOGFILE, (char *)buffer, tpos,
                                     2 * sizeof(afs_int32));
                if (code != 2 * sizeof(afs_int32))
                    return UBADLOG;
-               code = (*adbase->setlabel) (adbase, 0, (ubik_version *)buffer);
+               version.epoch = ntohl(buffer[0]);
+               version.counter = ntohl(buffer[1]);
+               code = (*adbase->setlabel) (adbase, 0, &version);
                if (code)
                    return code;
+               ubik_print("Successfully replayed log for interrupted "
+                          "transaction; db version is now %ld.%ld\n",
+                          (long) version.epoch, (long) version.counter);
                logIsGood = 1;
                break;          /* all done now */
            } else if (opcode == LOGTRUNCATE) {
@@ -447,7 +457,7 @@ urecovery_Interact(void *dummy)
     struct ubik_server *bestServer = NULL;
     struct ubik_server *ts;
     int dbok, doingRPC, now;
-    afs_int32 lastProbeTime, lastDBVCheck;
+    afs_int32 lastProbeTime;
     /* if we're the sync site, the best db version we've found yet */
     static struct ubik_version bestDBVersion;
     struct ubik_version tversion;
@@ -460,14 +470,13 @@ urecovery_Interact(void *dummy)
     char hoststr[16];
 #ifndef OLD_URECOVERY
     char pbuffer[1028];
-    int flen, fd = -1;
+    int fd = -1;
     afs_int32 pass;
 #endif
 
     /* otherwise, begin interaction */
     urecovery_state = 0;
     lastProbeTime = 0;
-    lastDBVCheck = 0;
     while (1) {
        /* Run through this loop every 4 seconds */
        tv.tv_sec = 4;
@@ -556,7 +565,6 @@ urecovery_Interact(void *dummy)
                 */
                urecovery_state &= ~UBIK_RECHAVEDB;
            }
-           lastDBVCheck = FT_ApproxTime();
            urecovery_state |= UBIK_RECFOUNDDB;
            urecovery_state &= ~UBIK_RECSENTDB;
        }
@@ -620,7 +628,6 @@ urecovery_Interact(void *dummy)
                goto FetchEndCall;
            }
 #ifndef OLD_URECOVERY
-           flen = length;
            afs_snprintf(pbuffer, sizeof(pbuffer), "%s.DB%s%d.TMP", ubik_dbase->pathName, (file<0)?"SYS":"", (file<0)?-file:file);
            fd = open(pbuffer, O_CREAT | O_RDWR | O_TRUNC, 0600);
            if (fd < 0) {