[hotfix] rocksdb log rename old.LOG.* when checkpoint and destroy_db

src/ctrip_swap_iter.c

@@ -289,6 +289,7 @@ rocksIter *rocksCreateIter(rocks *rocks, redisDb *db) {
         if (rocks->snapshot) rocksdb_readoptions_set_snapshot(rocks->iter_ropts, rocks->snapshot);
         rocksdb_options_t* cf_opts[CF_COUNT], *db_opts;
         db_opts = rocksdb_options_create_copy(rocks->db_opts);
+        rocksdb_options_set_db_log_dir(db_opts, "");
         /* disable background jobs since child have no background threads */
         rocksdb_options_set_max_background_jobs(db_opts, 0);
         rocksdb_options_set_max_background_compactions(db_opts, 0);

src/ctrip_swap_rocks.c

@@ -425,7 +425,10 @@ void rocksReleaseCheckpoint(rocks *rocks) {
         serverLog(LL_NOTICE, "[rocks] releasing checkpoint in (%s).", server.rocksdb_checkpoint_dir);
         rocksdb_checkpoint_object_destroy(server.rocksdb_checkpoint);
         server.rocksdb_checkpoint = NULL;
-        rocksdb_destroy_db(rocks->db_opts, server.rocksdb_checkpoint_dir, &err);
+        rocksdb_options_t *db_opts = rocksdb_options_create_copy(rocks->db_opts);
+        rocksdb_options_set_db_log_dir(db_opts, "");
+        rocksdb_destroy_db(db_opts, server.rocksdb_checkpoint_dir, &err);
+        rocksdb_options_destroy(db_opts);
         if (err != NULL) {
             serverLog(LL_WARNING, "[rocks] destory db fail: %s", server.rocksdb_checkpoint_dir);
         }

tests/swap/integration/rocksdb_log_rotate.tcl

@@ -0,0 +1,144 @@
+
+proc do_fullsync {slave master_host master_port} {
+    $slave slaveof no one
+    $slave set testing.fullsync 1
+    after 100
+    $slave slaveof $master_host $master_port
+    wait_for_sync $slave
+}
+
+
+proc count_old_logs {logs_dir} {
+    return [llength [glob -nocomplain "$logs_dir/*_LOG.old.*"]]
+}
+
+
+proc get_log_file {logs_dir} {
+    set list [glob -nocomplain "$logs_dir/*_LOG"]
+    if {[llength $list] != 1} {
+        return ""
+    }
+    return [lindex $list 0]
+}
+
+
+start_server {tags {"swap rocksdb-log-rotate"} overrides {
+    swap-repl-rordb-sync yes
+}} {
+    start_server {overrides {
+        swap-repl-rordb-sync yes
+    }} {
+        set master      [srv -1 client]
+        set master_host [srv -1 host]
+        set master_port [srv -1 port]
+        set master_dir  [lindex [$master config get dir] 1]
+        set slave       [srv 0 client]
+        set slave_dir  [lindex [$slave config get dir] 1]
+
+
+        set master_logs_dir  "$master_dir/logs.rocks"
+        set slave_logs_dir  "$slave_dir/logs.rocks"
+
+        $master mset k1 v1 k2 v2 k3 v3
+
+        test {rocksdb log rotate: step1 ror fullsync logs.rocks/LOG exist} {
+            $master set k1 v1
+            $master set k2 v2
+
+            $slave slaveof $master_host $master_port
+            wait_for_sync $slave
+
+            wait_for_condition 50 100 {
+                [get_log_file $master_logs_dir] != ""
+            } else {
+                fail "full sync logs.rocks/LOG exist fail"
+            }
+
+            assert {[file size [get_log_file $master_logs_dir]] > 1}
+            assert {[file size [get_log_file $slave_logs_dir]] > 1}
+        }
+
+        test {rocksdb log rotate:step2 ror fullsync slave LOG.old.*  count<=12} {
+
+
+            for {set i 0} {$i < 13} {incr i} {
+                do_fullsync $slave $master_host $master_port
+                wait_for_condition 50 200 {
+                    [status $master sync_full] == [expr {$i + 2}]
+                } else {
+                    fail "sync_full not updated"
+                }
+            }
+
+            after 500
+
+            assert {[get_log_file $master_logs_dir] != ""}
+
+            assert {[file size [get_log_file $master_logs_dir]] > 1}
+            assert {[count_old_logs $master_logs_dir] == 0}
+            assert {[file size [get_log_file $slave_logs_dir]] > 1}
+            assert {[count_old_logs $slave_logs_dir] == 11}
+        }
+
+
+    }
+}
+
+
+start_server {tags {"swap rocksdb-log-rotate rdb fullsync"} overrides {
+    swap-repl-rordb-sync no
+}} {
+    start_server {overrides {
+        swap-repl-rordb-sync no
+    }} {
+        set master      [srv -1 client]
+        set master_host [srv -1 host]
+        set master_port [srv -1 port]
+        set master_dir  [lindex [$master config get dir] 1]
+        set slave       [srv 0 client]
+        set slave_dir  [lindex [$slave config get dir] 1]
+
+
+
+        set master_logs_dir  "$master_dir/logs.rocks"
+        set slave_logs_dir  "$slave_dir/logs.rocks"
+
+        $master mset k1 v1 k2 v2 k3 v3
+
+        test {rocksdb log rotate: step1 rdb fullsync logs.rocks/LOG exist} {
+            $master set k1 v1
+            $master set k2 v2
+
+            $slave slaveof $master_host $master_port
+            wait_for_sync $slave
+
+            wait_for_condition 50 100 {
+                [get_log_file $master_logs_dir] != ""
+            } else {
+                fail "full sync logs.rocks/LOG exist fail"
+            }
+
+            assert {[file size [get_log_file $master_logs_dir]] > 1}
+            assert {[file size [get_log_file $slave_logs_dir]] > 1}
+        }
+
+        test {rocksdb log rotate:step2 rdb fullsync slave LOG.old.*  count == 0} {
+
+            for {set i 0} {$i < 13} {incr i} {
+                do_fullsync $slave $master_host $master_port
+                wait_for_condition 50 200 {
+                    [status $master sync_full] == [expr {$i + 2}]
+                } else {
+                    fail "sync_full not updated"
+                }
+            }
+
+            assert {[get_log_file $master_logs_dir] != ""}
+
+            assert {[file size [get_log_file $master_logs_dir]] > 1}
+            assert {[count_old_logs $master_logs_dir] == 0}
+            assert {[file size [get_log_file $slave_logs_dir]] > 1}
+            assert {[count_old_logs $slave_logs_dir] == 0}
+        }
+    }
+}

tests/swap/unit/ltrim_check.tcl

@@ -20,7 +20,7 @@ start_server {tags {"list" "repl"} overrides {save ""}} {
         $slave  config set swap-debug-evict-keys -1
 
         # Panic on replication error to detect the divergence immediately.
-        $slave config set propagation-error-behavior panic-on-replicas
+        # $slave config set propagation-error-behavior panic-on-replicas
 
         test {swap-list LTRIM cold-segment repl divergence: no panic and master/replica consistent} {
             set num_keys 4

tests/test_helper.tcl

@@ -122,6 +122,7 @@ set ::disk_tests {
 	swap/integration/expire_evict
 	swap/integration/swap_load
 	swap/integration/persist
+    swap/integration/rocksdb_log_rotate
 	swap/unit/swap_mode
 	swap/unit/absent_cache
 	swap/unit/dbsize