From 9ddab592fb2bbeacf4722962fac369b08e0ebccc Mon Sep 17 00:00:00 2001
From: Yuanjia Zhang <zhangyuanjia@pingcap.com>
Date: Wed, 3 Jul 2019 13:20:27 +0800
Subject: [PATCH] store/tikv: Make maxTxnTimeUse a configurable value (#10480)

---
 config/config.go           |  3 +++
 config/config.toml.example |  5 +++++
 store/tikv/2pc.go          | 36 ++++++++++++++++++++----------------
 3 files changed, 28 insertions(+), 16 deletions(-)

diff --git a/config/config.go b/config/config.go
index f8248398f64cb..f690f360a40bb 100644
--- a/config/config.go
+++ b/config/config.go
@@ -264,6 +264,8 @@ type TiKVClient struct {
 	GrpcKeepAliveTimeout uint `toml:"grpc-keepalive-timeout" json:"grpc-keepalive-timeout"`
 	// CommitTimeout is the max time which command 'commit' will wait.
 	CommitTimeout string `toml:"commit-timeout" json:"commit-timeout"`
+	// MaxTxnTimeUse is the max time a Txn may use (in seconds) from its startTS to commitTS.
+	MaxTxnTimeUse uint `toml:"max-txn-time-use" json:"max-txn-time-use"`
 }
 
 // Plugin is the config for plugin
@@ -357,6 +359,7 @@ var defaultConf = Config{
 		GrpcKeepAliveTime:    10,
 		GrpcKeepAliveTimeout: 3,
 		CommitTimeout:        "41s",
+		MaxTxnTimeUse:        590,
 	},
 	Binlog: Binlog{
 		WriteTimeout: "15s",
diff --git a/config/config.toml.example b/config/config.toml.example
index 8c04c46cde3ca..4545dfdf77a6e 100644
--- a/config/config.toml.example
+++ b/config/config.toml.example
@@ -253,6 +253,11 @@ grpc-keepalive-timeout = 3
 # max time for commit command, must be twice bigger than raft election timeout.
 commit-timeout = "41s"
 
+# The max time a Txn may use (in seconds) from its startTS to commitTS.
+# We use it to guarantee GC worker will not influence any active txn. Please make sure that this
+# value is less than gc_life_time - 10s.
+max-txn-time-use = 590
+
 [txn-local-latches]
 # Enable local latches for transactions. Enable it when
 # there are lots of conflicts between transactions.
diff --git a/store/tikv/2pc.go b/store/tikv/2pc.go
index 1e4ed7f2bf9c8..4a23632366758 100644
--- a/store/tikv/2pc.go
+++ b/store/tikv/2pc.go
@@ -23,6 +23,7 @@ import (
 	"github.com/pingcap/errors"
 	pb "github.com/pingcap/kvproto/pkg/kvrpcpb"
 	"github.com/pingcap/parser/terror"
+	"github.com/pingcap/tidb/config"
 	"github.com/pingcap/tidb/kv"
 	"github.com/pingcap/tidb/metrics"
 	"github.com/pingcap/tidb/sessionctx/binloginfo"
@@ -79,6 +80,11 @@ type twoPhaseCommitter struct {
 	connID   uint64 // connID is used for log.
 	cleanWg  sync.WaitGroup
 	detail   *execdetails.CommitDetails
+
+	// The max time a Txn may use (in ms) from its startTS to commitTS.
+	// We use it to guarantee GC worker will not influence any active txn. The value
+	// should be less than GC life time.
+	maxTxnTimeUse uint64
 }
 
 // newTwoPhaseCommitter creates a twoPhaseCommitter.
@@ -159,20 +165,23 @@ func newTwoPhaseCommitter(txn *tikvTxn, connID uint64) (*twoPhaseCommitter, erro
 		return nil, errors.Trace(err)
 	}
 
+	// Convert from sec to ms
+	maxTxnTimeUse := uint64(config.GetGlobalConfig().TiKVClient.MaxTxnTimeUse) * 1000
 	commitDetail := &execdetails.CommitDetails{WriteSize: size, WriteKeys: len(keys)}
 	metrics.TiKVTxnWriteKVCountHistogram.Observe(float64(commitDetail.WriteKeys))
 	metrics.TiKVTxnWriteSizeHistogram.Observe(float64(commitDetail.WriteSize))
 	return &twoPhaseCommitter{
-		store:     txn.store,
-		txn:       txn,
-		startTS:   txn.StartTS(),
-		keys:      keys,
-		mutations: mutations,
-		lockTTL:   txnLockTTL(txn.startTime, size),
-		priority:  getTxnPriority(txn),
-		syncLog:   getTxnSyncLog(txn),
-		connID:    connID,
-		detail:    commitDetail,
+		store:         txn.store,
+		txn:           txn,
+		startTS:       txn.StartTS(),
+		keys:          keys,
+		mutations:     mutations,
+		lockTTL:       txnLockTTL(txn.startTime, size),
+		priority:      getTxnPriority(txn),
+		syncLog:       getTxnSyncLog(txn),
+		connID:        connID,
+		detail:        commitDetail,
+		maxTxnTimeUse: maxTxnTimeUse,
 	}, nil
 }
 
@@ -593,11 +602,6 @@ func (c *twoPhaseCommitter) cleanupKeys(bo *Backoffer, keys [][]byte) error {
 	return c.doActionOnKeys(bo, actionCleanup, keys)
 }
 
-// The max time a Txn may use (in ms) from its startTS to commitTS.
-// We use it to guarantee GC worker will not influence any active txn. The value
-// should be less than `gcRunInterval`.
-const maxTxnTimeUse = 590000
-
 func (c *twoPhaseCommitter) executeAndWriteFinishBinlog(ctx context.Context) error {
 	err := c.execute(ctx)
 	if err != nil {
@@ -677,7 +681,7 @@ func (c *twoPhaseCommitter) execute(ctx context.Context) error {
 		return errors.Trace(err)
 	}
 
-	if c.store.oracle.IsExpired(c.startTS, maxTxnTimeUse) {
+	if c.store.oracle.IsExpired(c.startTS, c.maxTxnTimeUse) {
 		err = errors.Errorf("conn%d txn takes too much time, txnStartTS: %d, comm: %d",
 			c.connID, c.startTS, c.commitTS)
 		return errors.Annotate(err, txnRetryableMark)