TD-1949

src/inc/twal.h

@@ -59,6 +59,7 @@ int32_t walAlter(twalh pWal, SWalCfg *pCfg);
 void    walStop(twalh);
 void    walClose(twalh);
 int32_t walRenew(twalh);
+void    walRemoveOldFiles(twalh);
 int32_t walWrite(twalh, SWalHead *);
 void    walFsync(twalh, bool forceFsync);
 int32_t walRestore(twalh, void *pVnode, FWalWrite writeFp);

src/vnode/src/vnodeMain.c

@@ -583,6 +583,7 @@ static int vnodeProcessTsdbStatus(void *arg, int status) {
 
   if (status == TSDB_STATUS_COMMIT_OVER) {
     vDebug("vgId:%d, commit over, fver:%" PRIu64 " vver:%" PRIu64, pVnode->vgId, pVnode->fversion, pVnode->version);
+    walRemoveOldFiles(pVnode->wal);
     return vnodeSaveVersion(pVnode);
   }
 

src/wal/src/walMgmt.c

@@ -135,7 +135,7 @@ void walClose(void *handle) {
       if (remove(pWal->name) < 0) {
         wError("vgId:%d, wal:%p file:%s, failed to remove", pWal->vgId, pWal, pWal->name);
       } else {
-        wDebug("vgId:%d, wal:%p file:%s, it is removed", pWal->vgId, pWal, pWal->name);
+        wInfo("vgId:%d, wal:%p file:%s, it is removed", pWal->vgId, pWal, pWal->name);
       }
     }
   } else {

src/wal/src/walWrite.c

@@ -58,24 +58,32 @@ int32_t walRenew(void *handle) {
     wDebug("vgId:%d, file:%s, it is created", pWal->vgId, pWal->name);
   }
 
-  if (pWal->keep != TAOS_WAL_KEEP) {
-    // remove the oldest wal file
-    int64_t oldFileId = -1;
-    if (walGetOldFile(pWal, pWal->fileId, WAL_FILE_NUM, &oldFileId) == 0) {
-      char walName[WAL_FILE_LEN] = {0};
-      snprintf(walName, sizeof(walName), "%s/%s%" PRId64, pWal->path, WAL_PREFIX, oldFileId);
-
-      if (remove(walName) < 0) {
-        wError("vgId:%d, file:%s, failed to remove since %s", pWal->vgId, walName, strerror(errno));
-      } else {
-        wDebug("vgId:%d, file:%s, it is removed", pWal->vgId, walName);
-      }
+  pthread_mutex_unlock(&pWal->mutex);
+
+  return code;
+}
+
+void walRemoveOldFiles(void *handle) {
+  SWal *pWal = handle;
+  if (pWal == NULL) return;
+  if (pWal->keep == TAOS_WAL_KEEP) return;
+
+  pthread_mutex_lock(&pWal->mutex);
+
+  // remove the oldest wal file
+  int64_t oldFileId = -1;
+  if (walGetOldFile(pWal, pWal->fileId, WAL_FILE_NUM, &oldFileId) == 0) {
+    char walName[WAL_FILE_LEN] = {0};
+    snprintf(walName, sizeof(walName), "%s/%s%" PRId64, pWal->path, WAL_PREFIX, oldFileId);
+
+    if (remove(walName) < 0) {
+      wError("vgId:%d, file:%s, failed to remove since %s", pWal->vgId, walName, strerror(errno));
+    } else {
+      wInfo("vgId:%d, file:%s, it is removed", pWal->vgId, walName);
     }
   }
 
   pthread_mutex_unlock(&pWal->mutex);
-
-  return code;
 }
 
 int32_t walWrite(void *handle, SWalHead *pHead) {

tests/script/general/wal/kill.sim

+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+
+print ============== deploy
+system sh/exec.sh -n dnode1 -s start 
+sleep 3001
+sql connect
+
+sql create database d1
+sql use d1
+
+sql create table t1 (ts timestamp, i int) 
+sql insert into t1 values(now, 1);
+
+print ===============  step3
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL
+sleep 3000
+
+print ===============  step4
+system sh/exec.sh -n dnode1 -s start -x SIGKILL
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL
+sleep 3000
+
+print ===============  step5
+system sh/exec.sh -n dnode1 -s start -x SIGKILL
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL
+sleep 3000
+
+print ===============  step6
+system sh/exec.sh -n dnode1 -s start -x SIGKILL
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL
+sleep 3000
+
+print ===============  step7
+system sh/exec.sh -n dnode1 -s start -x SIGKILL
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL
+sleep 3000
+
+print ===============  step8
+system sh/exec.sh -n dnode1 -s start -x SIGKILL
+sleep 3000
+sql select * from t1;
+print rows: $rows
+if $rows != 1 then
+  return -1
+endi
+system sh/exec.sh -n dnode1 -s stop -x SIGKILL