Uploaded image for project: 'WiredTiger'
  1. WiredTiger
  2. WT-10007

Fix race when dropping local file

XMLWordPrintableJSON

      There is a race between a reader accessing the local store copy of a tiered object at the same time that we decide to delete that file. We drop a local file in  __tier_storage_remove_local():

              if (__wt_handle_is_open(session, object)) {
            __wt_verbose(session, WT_VERB_TIERED, "REMOVE_LOCAL: %s in USE, queue again", object);
            WT_STAT_CONN_INCR(session, local_objects_inuse);
            /* COMMENT ELIDED */
            WT_ASSERT(session, entry->tiered != NULL && entry->tiered->bstorage != NULL);
            entry->op_val = now + entry->tiered->bstorage->retain_secs;
            __wt_tiered_push_work(session, entry);
        } else {
            __wt_verbose(session, WT_VERB_TIERED, "REMOVE_LOCAL: actually remove %s", object);
            WT_STAT_CONN_INCR(session, local_objects_removed);
            WT_ERR(__wt_fs_remove(session, object, false));

      There is nothing here that prevents a file handle from being re-opened after the check in __wt_handle_is_open() and before the file remove in __wt_fs_remove(). If I insert a sleep before the remove and add some read activity to test_tiered08 I can trigger this exact race.

            Assignee:
            sue.loverso@mongodb.com Susan LoVerso
            Reporter:
            keith.smith@mongodb.com Keith Smith
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated:
              Resolved: