The Seata Parameter Configuration

Attention attribute (see all attributes for detailed description)

Server side	client side
registry.type	registry.type
config.type	config.type
#store.mode=db requires the following configuration	service.vgroupMapping.my* test* tx_ group
store.db.driverClassName	service.default. grouplist
store.db.url	service. disableGlobalTransaction
store.db.user
store.db.password
#store.mode=Redis requires the following configurations
store.redis.host
store.redis.port
store.redis.database
store.redis.password
#The following configurations are required for "store.mode=raft"
server.raft.group
server.raft.server-addr
server.raft.snapshot-interval

All attributes

Public sector

key	desc	remark	change record
transport.type	Socket communication mode	TCP, UNIX* DOMAIN* SOCKET, default TCP
transport. server	socket channel type	NIO, NATIVE (select KQueue or Epoll according to the operating system type and socket communication mode. Note that Windows only supports NIO, and NATIVE mode will throw an exception)
transport. enableTmClientBatchSendRequest	TM Batch Send Request Message Switch	Default false	new in 1.5.1 version
transport. enableRmClientBatchSendRequest	RM Batch Send Request Message Switch	Default true	new in 1.5.1 version
transport. enableTcServerBatchSendResponse	TC Batch Send Reply Message Switch	Default false	new in 1.5.1 version
transport.rpcRmRequestTimeout	RM sending request timeout	30 seconds by default	new in 1.5.1 version
transport.rpcTmRequestTimeout	TM sending request timeout	30 seconds by default	new in 1.5.1 version
transport.rpcTcRequestTimeout	TC sending request timeout	30 seconds by default	new in 1.5.1 version
transport.threadFactory. bossThreadSize	Netty communication model Boss group threads	Default 1
transport.threadFactory. workerThreadSize	Netty communication model Worker group threads	The number of threads can be configured or the number of threads in a specific thread working mode can be selected. There are four default working modes of threads: Auto (2 _ CPU cores+1), Pin (CPU cores, applicable to computing intensive tasks), BusyPin (CPU cores+1, applicable to computing intensive and memory limited scenarios) Default (2 _ CPU cores, applicable to IO intensive tasks), the default value is Default mode
transport. shutdown. wait	Time to wait for service offline before the Netty thread pool on the server is closed	3 seconds by default
transport.serialization	Client and server communication codec method	seata (ByteBuf), protobuf, kryo, hessian, default seata
transport.compressor	Compression method of communication data between client and server	none, gzip, zip, sevenz, bzip2, lz4, deflater, zstd, default none	Before 1.2.0: gzip 1.2.0: zip, sevenz, bzip2 1.3.0: lz4 1.4.1: deflater 1.5.1: zstd
transport. heartbeat	The heartbeat detection switch for client server communication	The default value is true
Registry.type	Registry type	Default file, supports file, nacos, redis, eureka, zk, consumer, etcd3, sofa, and custom	1.6.0 Server supports simultaneous registration to multiple registries, separating registry names with commas
Config.type	Configuration center type	default file, supporting file, nacos, apollo, zk, consult, etcd3, springcloud, custom

Server side

key	desc	remark	change record
server.undo.logSaveDays	undo retention days	7 days by default, log_ Status=1 (Appendix 3) and undo not normally cleaned
server.undo.logDeletePeriod	undo Cleanup Thread Interval	86,400,000 by default, in milliseconds
server.maxCommitRetryTimeout	Timeout duration of two-phase commit retry	Unit: ms, s, m, h, d, corresponding to ms, s, min, h, day, default ms. The default value of - 1 means unlimited retries. Formula: timeout>=now globalTransactionBeginTime, true means no more retries after timeout (Note: no retries will be made after the timeout is reached, and there is a risk of data inconsistency, unless the business can calibrate the data itself, otherwise use with caution)
server.maxRollbackRetryTimeout	Phase2 rollback retry timeout duration	Same as commit
server.recovery.committingRetryPeriod	Phase2 commit unfinished status Global transaction retry commit thread interval	1000 by default, in milliseconds
server.recovery.asyncCommittingRetryPeriod	Phase2 asynchronous submission status Retry submission thread interval	1000 by default, in milliseconds
server.recovery.rollbackRetryPeriod	Phase2 rollback status Retry rollback thread interval	1000 by default, in milliseconds
server.recovery.timeoutRetryPeriod	Timeout status detection retry thread interval	1000 by default, in milliseconds. If timeout is detected, put the global transaction into the rollback session manager
server.rollbackRetryTimeoutUnlockEnable	Whether to release the lock after the two-phase rollback timeout	False by default
server.distributedLockExpireTime	Server side transaction management global lock timeout	10000 by default, in milliseconds	new in 1.5.1 version
server.server.xaerNotaRetryTimeout	Retry timeout to prevent XA branch transactions from hanging	60000 by default, in milliseconds	new in 1.5.1 version
server.session.branchAsyncQueueSize	Branch transaction session asynchronously deletes the thread pool queue size	5000 by default	new in 1.5.1 version
server.session.enableBranchAsyncRemove	Asynchronous deletion switch of branch transaction session	Default false	new in 1.5.1 version
server.enableParallelRequestHandle	Parallel processing switch for batch request messages	Default false	new in 1.5.2 version
server.enableParallelHandleBranch	Two-phase Parallel Deployment Switch	false	new in 2.0.0 version
server.raft.group	In the Raft storage mode, the transaction group corresponding to the group and client should be matched with the value. For example, service.vgroup-mapping.default_tx_group=default.	New in 2.0.0 version
server.raft.server-addr	The Raft cluster list is as follows: 192.168.0.111:9091, 192.168.0.112:9091, 192.168.0.113:9091		New in 2.0.0 version
server.raft.snapshot-interval	How often to take a memory snapshot? The state machine is paused every time a snapshot is taken, but this can improve recovery speed when stopping	默认 600 秒	New in 2.0.0 version
server.raft.apply-batch	Accumulate batches of tasks and submit them to the leader.	32	New in 2.0.0 version
server.raft.max-append-bufferSize	Maximum size of the Raft log storage buffer	262144	New in 2.0.0 version
server.raft.max-replicator-inflight-msgs	The maximum number of in-flight requests when pipeline requests are enabled	256	New in 2.0.0 version
server.raft.disruptor-buffer-size	The size of the internal disruptor buffer may need to be increased appropriately in scenarios with high write throughput	16384	New in 2.0.0 version
server.raft.election-timeout-ms	How long does it take to start a new election when there is no leader heartbeat for a long time	1000	New in 2.0.0 version
server.raft.reporter-enabled	Raft monitor enable	false	New in 2.0.0 version
server.raft.reporter-initial-delay	Monitor output interval	60	New in 2.0.0 version
server.raft.serialization	raft log serialization type	jackson	New in 2.0.0 version
server.raft.compressor	The compression methods supported for Raft log and snapshot are gzip, zstd, and lz4	none	New in 2.0.0 version
server.raft.sync	Synchronous flushing of Raft log to disk	true	New in 2.0.0 version
store.mode	Transaction session information storage mode	file Local file (HA is not supported), db database, redis, raft (HA is supported)	The 1.5.1 version uses lock and session for separate storage
store.lock.mode	Transaction lock information storage method	file Local file (HA is not supported), db database, Redis (HA is supported); When the configuration is empty, take the store Mode configuration item value. It is not allowed to specify separately in Raft mode	New in 1.5.1 version, and session and lock can be stored separately
store.session.mode	Transaction echo information storage method	file Local file (HA is not supported), db database, Redis (HA is supported); When the configuration is empty, take the store Mode configuration item value. It is not allowed to specify separately in Raft mode	New in 1.5.1 version, and session and lock can be stored separately
store.publicKey	Public key for storing and decrypting passwords in db or Redis		supported in version 1.4.2
store.file.dir	file mode file storage folder name	default sessionStore
store.file.maxBranchSessionSize	File mode file storage branch session maximum bytes	16384 (16kb) by default, in bytes
store.file.maxGlobalSessionSize	File mode file stores the maximum number of global session bytes	512b by default, in bytes
store.file.fileWriteBufferCacheSize	File mode file storage buffer maximum cache size	16384 (16kb) by default, in bytes. When the amount of data such as session written is greater than this value, an exception will be thrown
store.file.flushDiskMode	file mode file storage flushing policy	default async, optional sync
store.file.sessionReloadReadSize	File mode File storage The maximum number of session or lock keys recovered from the backup file after the server node restarts	100 by default
store.db.datasource	db mode data source type	dbcp, druid, hikari; No default value, store Must be specified when mode=db
store.db.dbType	db mode database type	mysql, oracle, db2, sqlserver, sybaee, h2, sqlite, access, postgresql, oceanbase; No default value, store Mode=db must be specified
store.db.driverClassName	db mode database driver	store Must be specified when mode=db
store.db.url	db mode database url	store When mode=db, it must be specified. When using MySQL as the data source, it is recommended to add 'rewriteBatchedStatements=true' to the connection parameters (see Appendix 7 for detailed reasons)
store.db.user	db mode database account	store Must be specified when mode=db
store.db.password	db mode database account password	store Must be specified when mode=db
store.db.minConn	initial connections of db mode database	1 by default
store.db.maxConn	maximum number of connections to database in db mode	20 by default
store.db.maxWait	the maximum waiting time for db mode to obtain a connection	5000 by default, in milliseconds
store.db.globalTable	db mode global transaction table name	default global_ table
store.db.branchTable	db mode branch transaction table name	default branch_ table
store.db.LockTable	db mode global lock table name	default lock_ table
store.db.queryLimit	the maximum number of global transactions queried in db mode at one time	100 by default
store.db.distributedLockTable	db mode Sever side transaction management global lock storage table name	default distributed_ Lock. Under multiple server clusters, only one server is guaranteed to handle the submission or rollback at the same time	New in 1.5.1
store.redis.mode	redis mode	default single, optional sentinel	New sentinel mode in 1.4.2
store.redis.single.host	The Redis host in stand-alone mode is compatible with versions earlier than 1.4.2. If this configuration is empty, select store.redis Host as a configuration item	New in 1.4.2
store.redis.single.port	The port of Redis in stand-alone mode is compatible with versions earlier than 1.4.2. If this configuration is empty, select store.redis Port as a configuration item	New in 1.4.2
store.redis.sentinel.masterName	Redis master database name in sentinel mode		New in 1.4.2
store.redis.sentinel.sentinelHosts	sentinel hosts in sentinel mode	Multiple hosts are separated by commas	New in 1.4.2
store.redis.host	redis mode IP	Default 127.0.0.1	Version 1.4.2 deprecated
store.redis.port	redis mode port	6379 by default	Version 1.4.2 is deprecated
store.redis.maxConn	maximum connections in Redis mode	10 by default
store.redis.minConn	minimum connections in Redis mode	1 by default
store.redis.database	redis mode default library	default 0
store.redis.password	redis mode password (optional)	null by default
store.redis.queryLimit	The maximum number of Redis queries at a time	100 by default
store.redis.type	Redis operation type: lua, pippline	pippline
metrics.enabled	Whether to enable Metrics	False is off by default. In the false state, all Metrics related components will not be initialized to minimize the performance loss
metrics.registryType	Indicator registrar type	The indicator registrar type used by Metrics is a built-in compact (simple) implementation by default. Meters in this implementation only use a limited memory count, and the performance is high enough to meet most scenarios; Currently, only one indicator registrar can be set
metrics.exporterList	Index result Measurement data outputter list	default prometheus. Multiple outputters are separated by English commas, such as "prometheus, jmx". Currently, only the prometheus outputters are connected
metrics.exporterPrometheusPort	prometheus exporter client port number	9898 by default

Client side

key	desc	remark	change record
seata.enabled	whether to enable spring boot automatic assembly	true, false, (SSBS) special configuration, default true (Appendix 4)
seata.enableAutoDataSourceProxy=true	whether to enable automatic proxy of data source	True, false, data spring boot starter (SSBS) is a special configuration. SSBS will enable automatic proxy of data source by default and can be closed through this configuration item
seata.useJdkProxy=false	whether to use JDK proxy as the implementation method of data source automatic proxy	true, false, (SSBS) proprietary configuration, false by default, CGLIB as the implementation method of data source automatic proxy
transport.enableClientBatchSendRequest	whether to batch merge and send client transaction message requests	The default is true and false
client.log.exceptionRate	log exception output probability	100 by default, currently used for exception stack output in case of undo rollback failure, 1% probability output, rollback failure is basically dirty data, no need to output stack to occupy hard disk space
service.vgroupMapping.my_test_tx_Group	transaction group (Appendix 1)	my* test* tx_ Group refers to grouping, and the configuration item value is TC cluster name
service.default.grouplist	TC service list (Appendix 2)	Only used when the registry is file
service.disableGlobalTransaction	global transaction switch	False by default. false is on, true is off
client.tm.degradeCheck	demote switch	False by default. The business side automatically degradeCheck the seata transaction based on the number of consecutive errors (see Appendix 6 for details)
client.tm.degradeCheckAllowTimes	threshold value for reaching the standard of promotion and demotion	10 by default
client.tm.degradeCheckPeriod	service self check cycle	2000 by default, in ms Conduct a service self check every 2 seconds to detect
client.rm.reportSuccessEnable	whether to report phase I success	True, false, starting from version 1.1.0. The default is false true is used to keep the life cycle records of branch transactions complete. false can improve performance
client.rm.asyncCommitBufferLimit	asynchronous commit cache queue length	10000 by default. Phase II commit succeeded. RM asynchronously cleans the undo queue
client.rm.lock.retryInterval	retry interval for verifying or occupying the global lock	10 by default, in milliseconds
client.rm.lock.retryTimes	number of retries to verify or occupy the global lock	30 by default
client.rm.lock.retryPolicyBranchRollbackOnConflict	the lock policy when a branch transaction conflicts with other global rollback transactions	The default is true. Release the local lock first to allow rollback to succeed
client.rm.reportRetryCount	the number of TC retries for reporting the results in the first phase	5 by default	New in 1.4.1
client.rm.tableMetaCheckEnable	automatically refresh the table structure in the cache	Default false	New in 1.5.1
client.rm.tableMetaCheckerInterval	interval between scheduled refreshing of table structures in cache	60 seconds by default
client.rm.sagaBranchRegisterEnable	whether to enable saga branch registration	In Saga mode, the branch state is stored in the local database of the state machine and can be submitted or rolled back through the state machine. To improve performance, you can consider not registering Saga branches with the TC, but you need to consider the availability of the state machine. The default is false
client.rm.sagaJsonParser	data serialization method in saga mode	Fastjson by default, jackson by choice	New in 1.5.1
client.rm.tccActionInterceptorOrder	tcc interceptor order	default Ordered HIGHEST_ PRECEDENCE+1000 ensures that the interceptor is executed before the local transaction interceptor. You can also customize the execution order of the interceptor for tcc and business development	New in 1.5.1
client.tm.commitRetryCount	the number of TC retries for reporting the results of the first phase global submission	1 by default, it is recommended to be greater than 1
client.tm.rollbackRetryCount	the number of TC retries reported for the global rollback results in the first phase	1 by default, it is recommended to be greater than 1
client.tm.defaultGlobalTransactionTimeout	global transaction timeout	The default is 60 seconds. After TM detects a branch transaction timeout or TC detects that TM has not made a two-phase report timeout, it initiates a rollback of the branch transaction	New in 1.4.0
client.tm.interceptorOrder	TM global transaction interceptor order	Ordered by default HIGHEST_ PRECEDENCE+1000 ensures that the interceptor is executed before the local transaction interceptor. You can also customize the execution order of the interceptor for global transactions and business development	New in 1.5.1
client.undo.dataValidation	Phase2 rollback image verification	By default, true is enabled and false is disabled
client.undo.logSerialization	undo serialization method	default jackson
client.undo.logTable	user defined undo table name	Default undo_ log
client.undo.onlyCareUpdateColumns	only images of updated columns are generated	Default true
client.undo.compress.enable	undo log compression switch	default true	New in 1.4.1
client.undo.compress.type	undo log compression algorithm	Default zip. Options include NONE, GZIP, ZIP, SEVENZ, BZIP2, LZ4, DEFLATER, and ZSTD	New in 1.4.1
client.undo.compress.threshold	undo log compression threshold	The default value is 64k. Compression occurs when the compression switch is on and the undo log size exceeds the threshold value	New in 1.4.1
client.rm.sqlParserType	sql resolution type	default druid, optional antlr

synchronize parameters to the configuration center for use demo

Nacos

shell:

sh $\{SEATAPATH}/script/config-center/nacos/nacos-config.sh -h localhost -p 8848 -g SEATA_GROUP -t 5a3c7d6c-f497-4d68-a71a-2e5e3340b3ca

Parameter description:

-h: Host, the default is localhost

-p: Port, default 8848

-g: Configure grouping. The default value is' SEATA_ GROUP'

-t: Tenant information, corresponding to the namespace ID field of Nacos, the default value is null ''

Apollo

sh $\{SEATAPATH}/script/config-center/apollo/apollo-config.sh -h localhost -p 8070 -e DEV -a seata-server -c default -n application -d apollo -r apollo -t 3aa026fc8435d0fc4505b345b8fa4578fb646a2c

Parameter description:

-h: Host, the default is localhost

-p: Port, the default value is 8070

-e: Managed configuration environment, default value DEV

-a: AppId of the Namespace. The default value is seata server

-c: The name of the managed configuration cluster is generally passed in as default. If it is a special cluster, just pass in the name of the corresponding cluster. The default value is default

-n: The name of the namespace under management. If it is not in the properties format, it needs to add a suffix, such as sample.yml. The default value is application

-d: The creator of the item, in the format of domain account, that is, the user ID of the sso system

-r: Publisher, domain account. Note: If the namespace. lock. switch in Apollo ConfigDB. ServerConfig is set to true (false by default), the environment does not allow the publisher and editor to be the same person. So if the editor is zhangsan, the publisher can no longer be zhangsan.

-t: The Apollo administrator creates a third-party application at http://{portal_address}/open/manage.html. Before creating it, it is better to query whether the AppId has been created. After successful creation, a token will be generated

For details of the above parameters, see:

https://github.com/ctripcorp/apollo/wiki/Apollo%E5%BC%80%E6%94%BE%E5%B9%B3%E5%8F%B0

Consul

sh $\{SEATAPATH}/script/config-center/consul/consul-config.sh -h localhost -p 8500

Parameter description:

-h: host, the default is localhost

-p: port, the default value is 8500

Etcd3

sh $\{SEATAPATH}/script/config-center/etcd3/etcd3-config.sh -h localhost -p 2379

Parameter description:

-h: Host, the default is localhost

-p: Port, the default value is 2379

python:

python $\{SEATAPATH}/script/config-center/nacos/nacos-config.py localhost:8848

ZK

sh $\{SEATAPATH}/script/config-center/zk/zk-config.sh -h localhost -p 2181 -z "/Users/zhangchenghui/zookeeper-3.4.14"

Parameter description:

-h: host, the default is localhost

-p: port, the default value is 2181

-z: zk path

Appendix 1:

Description of transaction grouping.

1. What is a transaction group?

Transaction grouping is the resource logic of seata, similar to service instance. My in file.conf* test* tx_ A group is a transaction group.

2. How to find the back-end cluster through transaction grouping?

First, the program configures transaction grouping (txServiceGroup parameter of GlobalTransactionScanner construction method). The program will search for service.vgroupMapping through the user configured configuration center

Transaction grouping configuration item. The value of the configuration item obtained is the name of the TC cluster. The program that obtains the cluster name constructs the service name through a certain prefix+cluster name. The service name of each configuration center is different. Get the service name, go to the corresponding registry, pull the service list of the corresponding service name, and get the real TC service list of the back-end.

3. Why is the design so that the service name is not directly taken?

There is an additional layer to obtain the configuration of transaction groups to the mapping cluster. After this design, the transaction group can be used as the logical isolation unit of resources, which can quickly fail over when a failure occurs.

Appendix 2:

About the grouplist question.

1. When will the default.grouplist in file.conf be used?

It is used when registry.type=file. It is not read in other times.

2. Can multiple value lists be configured for default.grouplist?

Multiple can be configured, which means cluster, but when store When mode=file, an error will be reported. The reason is that the file storage mode does not provide synchronization of local files, so you need to use store.mode=db to share data between TC clusters through db

3. Is default.grouplist recommended?

Not recommended. For example, question 1, when registering It is used when type=file, which means that the real registry is not used here, and the health check mechanism without specific services cannot automatically remove the list when the tc is unavailable. It is recommended to use nacos, eureka, redis, zk, consumer, etcd3, and sofa. Registry. type=file or config The original purpose of type=file design is to enable users to quickly verify the seata service through direct connection without relying on a third-party registry or configuration center.

4. Why is the configuration in seata spring boot starter grouplist.default, that is, the default in file.conf Is the writing of grouplist just reversed?

Due to the syntax requirements of the spring boot configuration file itself, the default.grouplist in file.conf needs to be written as grouplist.default in this place. The effect is the same

Appendix 3:

log_ Status=1 is defensive. It is a global rollback request received, but it is uncertain whether the local transaction of a transaction branch has been executed. At this time, a piece of data with the same branch ID is inserted in advance, and the inserted false data is successful. If the local transaction continues to execute, the only index conflict will be reported for automatic rollback.

If the insertion is unsuccessful, it indicates that the local transaction has been completed, and the undo data is taken out for reverse rollback.

Appendix 4:

Whether to enable spring boot automatic assembly. If it is enabled, the integration of seata and spring boot will be automatically configured, including the automatic proxy of data source and initialization of GlobalTransactionScanner.

Note: New features of version 1.0 need to rely on seata spring boot starter.

Appendix 5:

The following annotations have been added to seata 1.1.0 to enable the automatic proxy function of data sources

@EnableAutoDataSourceProxy

attribute	desc	remark

|UseJdkProxy | Whether to use JDK proxy as the implementation method of data source automatic proxy | false, true, false by default, CGLIB as the implementation method of data source automatic proxy|

1. For the method of using seata spring boot starter, the data source automatic proxy is enabled by default. If you need to close it, please configure seata. enableAutoDataSourceProxy=false, which is true by default.

To switch the proxy implementation mode, please click seata UseJdkProxy=false. The default value is false. CGLIB is used as the implementation method of automatic proxy for data sources.

2. For the method of using seata all, please use @ EnableAutoDataSourceProxy to explicitly enable the automatic proxy function of the data source. If necessary, you can use the useJdkProxy attribute of the annotation to implement the proxy

Switch of. The default value is false. CGLIB is used as the implementation method of automatic proxy for data sources.

Appendix 6:

Introduction to the implementation of automatic service degradation policy:

First, decide whether to start the self-test thread by reading whether client.tm.degradeCheck is true. Then read degradeCheckAllowTimes and degradeCheckPeriod to confirm the threshold value and self-test cycle

Assume that degradeCheckAllowTimes=10, and degradeCheckPeriod=2000

Then a start and commit test will be performed every 2 seconds. If it fails, the number of consecutive failures will be recorded. If it succeeds, the number of consecutive failures will be cleared The continuous errors are accumulated by the user interface and the self checking thread until the number of continuous failures reaches the user's threshold, then the Seata distributed transaction is closed to avoid the user's own business being unavailable for a long time

On the contrary, if the current distributed transaction is closed, the self-test thread will continue to perform the self-test every 2 seconds until the number of consecutive successes reaches the threshold set by the user, then the Seata distributed transaction will be restored to use

Appendix 7:

In the store mode=db. Because seata inserts global locks in batches through the executeBatch of jdbc, according to the MySQL official website, when the rewriteBatchedStatements in the connection parameter is true, when executeBatch is executed and the operation type is insert, the jdbc driver will optimize the corresponding SQL to the form of 'insert into () values (), ()' to improve the performance of batch insert.

According to the actual test, when this parameter is set to true, the corresponding batch insert performance is more than 10 times of the original. Therefore, it is recommended to set this parameter to true when the data source is MySQL.