recover: finish pending trans3 commits when a recovery is finished.
[sahlberg/ctdb.git] / server / ctdb_vacuum.c
index 2cbee309186238879f40a1cd0391855a69479977..4aac302dca715ea1f89120fc079b145c2b919043 100644 (file)
@@ -18,7 +18,7 @@
 */
 
 #include "includes.h"
-#include "lib/events/events.h"
+#include "lib/tevent/tevent.h"
 #include "lib/tdb/include/tdb.h"
 #include "system/network.h"
 #include "system/filesys.h"
@@ -26,7 +26,7 @@
 #include "../include/ctdb_private.h"
 #include "db_wrap.h"
 #include "lib/util/dlinklist.h"
-#include "lib/events/events.h"
+#include "lib/tevent/tevent.h"
 #include "../include/ctdb_private.h"
 #include "../common/rb_tree.h"
 
@@ -36,7 +36,9 @@
 enum vacuum_child_status { VACUUM_RUNNING, VACUUM_OK, VACUUM_ERROR, VACUUM_TIMEOUT};
 
 struct ctdb_vacuum_child_context {
+       struct ctdb_vacuum_child_context *next, *prev;
        struct ctdb_vacuum_handle *vacuum_handle;
+       /* fd child writes status to */
        int fd[2];
        pid_t child_pid;
        enum vacuum_child_status status;
@@ -272,7 +274,7 @@ static int ctdb_vacuum_db(struct ctdb_db_context *ctdb_db, struct vacuum_data *v
 
                        data.dsize = talloc_get_size(vdata->list[i]);
                        data.dptr  = (void *)vdata->list[i];
-                       if (ctdb_send_message(ctdb, ctdb->vnn_map->map[i], CTDB_SRVID_VACUUM_FETCH, data) != 0) {
+                       if (ctdb_client_send_message(ctdb, ctdb->vnn_map->map[i], CTDB_SRVID_VACUUM_FETCH, data) != 0) {
                                DEBUG(DEBUG_ERR,(__location__ " Failed to send vacuum fetch message to %u\n",
                                         ctdb->vnn_map->map[i]));
                                return -1;              
@@ -516,6 +518,7 @@ static int update_tuning_db(struct ctdb_db_context *ctdb_db, struct vacuum_data
        struct vacuum_tuning_data tdata;
        struct vacuum_tuning_data *tptr;
        char *vac_dbname;
+       int flags;
 
        vac_dbname = talloc_asprintf(tmp_ctx, "%s/%s.%u",
                                     ctdb_db->ctdb->db_directory_state,
@@ -526,8 +529,10 @@ static int update_tuning_db(struct ctdb_db_context *ctdb_db, struct vacuum_data
                return -1;
        }
 
+       flags  = ctdb_db->ctdb->valgrinding ? TDB_NOMMAP : 0;
+       flags |= TDB_DISALLOW_NESTING;
        tune_tdb = tdb_open(vac_dbname, 0,
-                           TDB_DISALLOW_NESTING,
+                           flags,
                            O_RDWR|O_CREAT, 0600);
        if (tune_tdb == NULL) {
                DEBUG(DEBUG_ERR,(__location__ " Failed to create/open %s\n", TUNINGDBNAME));
@@ -568,12 +573,12 @@ static int update_tuning_db(struct ctdb_db_context *ctdb_db, struct vacuum_data
                                tdata.new_interval > ctdb_db->ctdb->tunable.vacuum_max_interval) {
                                tdata.new_interval = ctdb_db->ctdb->tunable.vacuum_min_interval;
                        }               
-                       DEBUG(DEBUG_ERR,("Decreasing vacuum interval %u -> %u for %s\n", 
+                       DEBUG(DEBUG_INFO,("Decreasing vacuum interval %u -> %u for %s\n", 
                                         tdata.last_interval, tdata.new_interval, ctdb_db->db_name));
                }
                tdata.last_interval = tdata.new_interval;
        } else {
-               DEBUG(DEBUG_ERR,(__location__ " Cannot find tunedb record for %s. Using default interval\n", ctdb_db->db_name));
+               DEBUG(DEBUG_DEBUG,(__location__ " Cannot find tunedb record for %s. Using default interval\n", ctdb_db->db_name));
                tdata.last_num_repack = freelist;
                tdata.last_num_empty = vdata->delete_count;
                tdata.last_interval = ctdb_db->ctdb->tunable.vacuum_default_interval;
@@ -631,6 +636,7 @@ static int ctdb_repack_db(struct ctdb_db_context *ctdb_db, TALLOC_CTX *mem_ctx)
        vdata->vacuum_limit = vacuum_limit;
        vdata->repack_limit = repack_limit;
        vdata->delete_tree = trbt_create(vdata, 0);
+       vdata->ctdb_db = ctdb_db;
        if (vdata->delete_tree == NULL) {
                DEBUG(DEBUG_ERR,(__location__ " Out of memory\n"));
                talloc_free(vdata);
@@ -681,6 +687,7 @@ static int get_vacuum_interval(struct ctdb_db_context *ctdb_db)
        char *vac_dbname;
        uint interval = ctdb_db->ctdb->tunable.vacuum_default_interval;
        struct ctdb_context *ctdb = ctdb_db->ctdb;
+       int flags;
 
        vac_dbname = talloc_asprintf(tmp_ctx, "%s/%s.%u", ctdb->db_directory, TUNINGDBNAME, ctdb->pnn);
        if (vac_dbname == NULL) {
@@ -689,11 +696,13 @@ static int get_vacuum_interval(struct ctdb_db_context *ctdb_db)
                return interval;
        }
 
+       flags  = ctdb_db->ctdb->valgrinding ? TDB_NOMMAP : 0;
+       flags |= TDB_DISALLOW_NESTING;
        tdb = tdb_open(vac_dbname, 0,
-                      TDB_DISALLOW_NESTING,
-                      O_RDWR|O_CREAT, 0644);
+                      flags,
+                      O_RDWR|O_CREAT, 0600);
        if (!tdb) {
-               DEBUG(DEBUG_ERR,("Unable to open/create database %s using default interval\n", vac_dbname));
+               DEBUG(DEBUG_ERR,("Unable to open/create database %s using default interval. Errno : %s (%d)\n", vac_dbname, strerror(errno), errno));
                talloc_free(tmp_ctx);
                return interval;
        }
@@ -737,6 +746,8 @@ static int vacuum_child_destructor(struct ctdb_vacuum_child_context *child_ctx)
                kill(child_ctx->child_pid, SIGKILL);
        }
 
+       DLIST_REMOVE(ctdb->vacuumers, child_ctx);
+
        event_add_timed(ctdb->ev, child_ctx->vacuum_handle,
                        timeval_current_ofs(get_vacuum_interval(ctdb_db), 0), 
                        ctdb_vacuum_event, child_ctx->vacuum_handle);
@@ -795,10 +806,17 @@ ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
        struct ctdb_db_context *ctdb_db = vacuum_handle->ctdb_db;
        struct ctdb_context *ctdb = ctdb_db->ctdb;
        struct ctdb_vacuum_child_context *child_ctx;
+       struct tevent_fd *fde;
        int ret;
 
-       /* we dont vacuum if we are in recovery mode */
-       if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE) {
+       /* we dont vacuum if we are in recovery mode, or db frozen */
+       if (ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE ||
+           ctdb->freeze_mode[ctdb_db->priority] != CTDB_FREEZE_NONE) {
+               DEBUG(DEBUG_INFO, ("Not vacuuming %s (%s)\n", ctdb_db->db_name,
+                                  ctdb->recovery_mode == CTDB_RECOVERY_ACTIVE ? "in recovery"
+                                  : ctdb->freeze_mode[ctdb_db->priority] == CTDB_FREEZE_PENDING
+                                  ? "freeze pending"
+                                  : "frozen"));
                event_add_timed(ctdb->ev, vacuum_handle, timeval_current_ofs(ctdb->tunable.vacuum_default_interval, 0), ctdb_vacuum_event, vacuum_handle);
                return;
        }
@@ -818,7 +836,7 @@ ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
                return;
        }
 
-       child_ctx->child_pid = fork();
+       child_ctx->child_pid = ctdb_fork(ctdb);
        if (child_ctx->child_pid == (pid_t)-1) {
                close(child_ctx->fd[0]);
                close(child_ctx->fd[1]);
@@ -835,7 +853,7 @@ ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
 
                DEBUG(DEBUG_INFO,("Vacuuming child process %d for db %s started\n", getpid(), ctdb_db->db_name));
        
-               if (switch_from_server_to_client(ctdb) != 0) {
+               if (switch_from_server_to_client(ctdb, "vacuum-%s", ctdb_db->db_name) != 0) {
                        DEBUG(DEBUG_CRIT, (__location__ "ERROR: failed to switch vacuum daemon into client mode. Shutting down.\n"));
                        _exit(1);
                }
@@ -855,23 +873,34 @@ ctdb_vacuum_event(struct event_context *ev, struct timed_event *te,
        child_ctx->status = VACUUM_RUNNING;
        child_ctx->start_time = timeval_current();
 
+       DLIST_ADD(ctdb->vacuumers, child_ctx);
        talloc_set_destructor(child_ctx, vacuum_child_destructor);
 
        event_add_timed(ctdb->ev, child_ctx,
                timeval_current_ofs(ctdb->tunable.vacuum_max_run_time, 0),
                vacuum_child_timeout, child_ctx);
 
-       DEBUG(DEBUG_INFO, (__location__ " Created PIPE FD:%d to child vacuum process\n", child_ctx->fd[0]));
+       DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to child vacuum process\n", child_ctx->fd[0]));
 
-       event_add_fd(ctdb->ev, child_ctx, child_ctx->fd[0],
-               EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
-               vacuum_child_handler,
-               child_ctx);
+       fde = event_add_fd(ctdb->ev, child_ctx, child_ctx->fd[0],
+                          EVENT_FD_READ, vacuum_child_handler, child_ctx);
+       tevent_fd_set_auto_close(fde);
 
        vacuum_handle->child_ctx = child_ctx;
        child_ctx->vacuum_handle = vacuum_handle;
 }
 
+void ctdb_stop_vacuuming(struct ctdb_context *ctdb)
+{
+       /* Simply free them all. */
+       while (ctdb->vacuumers) {
+               DEBUG(DEBUG_INFO, ("Aborting vacuuming for %s (%i)\n",
+                          ctdb->vacuumers->vacuum_handle->ctdb_db->db_name,
+                          (int)ctdb->vacuumers->child_pid));
+               /* vacuum_child_destructor kills it, removes from list */
+               talloc_free(ctdb->vacuumers);
+       }
+}
 
 /* this function initializes the vacuuming context for a database
  * starts the vacuuming events