Enable fail-fast test and other fixes.

go/sync2/consolidator.go

@@ -14,10 +14,6 @@ import (
 	"github.com/youtube/vitess/go/cache"
 )
 
-var (
-	waitError = fmt.Errorf("Error waiting for consolidation")
-)
-
 // Consolidator consolidates duplicate queries from executing simulaneously
 // and shares results between them.
 type Consolidator struct {
@@ -49,9 +45,7 @@ func (co *Consolidator) Create(query string) (r *Result, created bool) {
 	if r, ok := co.queries[query]; ok {
 		return r, false
 	}
-	// Preset the error. If there was an unexpected panic during the main
-	// query, then all those who waited will return the waitError.
-	r = &Result{consolidator: co, query: query, Err: waitError}
+	r = &Result{consolidator: co, query: query}
 	r.executing.Lock()
 	co.queries[query] = r
 	return r, true

go/vt/tabletserver/request_context.go

@@ -48,10 +48,6 @@ func (rqc *RequestContext) qFetch(logStats *SQLQueryStats, parsedQuery *sqlparse
 	q, ok := rqc.qe.consolidator.Create(string(sql))
 	if ok {
 		defer q.Broadcast()
-		// Wrap default error to TabletError
-		if q.Err != nil {
-			q.Err = NewTabletError(ErrFail, q.Err.Error())
-		}
 		waitingForConnectionStart := time.Now()
 		timeout, err := rqc.deadline.Timeout()
 		if err != nil {

test/vtgatev2_test.py

@@ -925,8 +925,6 @@ class TestFailures(unittest.TestCase):
     fail-fast (returning an appropriate error) without waiting around till the
     request deadline expires.
     """
-    # TODO(liguo): verify the upper bound for max RTT enable this test after bug fix
-    return
     try:
       tablet_type = 'replica'
       keyranges = [get_keyrange(shard_names[0])]
@@ -936,8 +934,8 @@ class TestFailures(unittest.TestCase):
 
       # Shutdown mysql and ensure tablet is in NOT_SERVING state
       utils.wait_procs([shard_0_replica.shutdown_mysql()])
-      utils.run_vtctl(['RunHealthCheck', shard_0_replica.tablet_alias, tablet_type])
       try:
+        get_rtt(KEYSPACE_NAME, query, tablet_type, keyranges)
         shard_0_replica.wait_for_vttablet_state('NOT_SERVING')
       except Exception, e:
         self.fail('unable to set tablet to NOT_SERVING state')
@@ -954,15 +952,14 @@ class TestFailures(unittest.TestCase):
       rt_times = []
       for async_result in async_results:
         rt_times.append(async_result.get())
-      # The true upper limit is 2 seconds (200ms *10 retries). To account for
+      # The true upper limit is 2 seconds (1s * 2 retries as in utils.py). To account for
       # network latencies and other variances, we keep an upper bound of 3 here.
       self.assertTrue(max(rt_times) < 3, 'at least one request did not fail-fast; round trip times: %s' % rt_times)
 
       # Restart tablet and put it back to SERVING state
       utils.wait_procs([shard_0_replica.start_mysql(),])
       shard_0_replica.kill_vttablet()
-      shard_0_replica.start_vttablet(wait_for_state='NOT_SERVING')
-      utils.run_vtctl(['RunHealthCheck', shard_0_replica.tablet_alias, tablet_type])
+      shard_0_replica.start_vttablet()
       shard_0_replica.wait_for_vttablet_state('SERVING')
     except Exception, e:
       logging.debug("failed with error %s, %s" % (str(e), traceback.print_exc()))