diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index d2c6e15..333fec7 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -1066,28 +1066,73 @@ OutputPluginWrite(ctx, true); Synchronous Replication Support for Logical Decoding + + Overview - - Logical decoding can be used to build - synchronous - replication solutions with the same user interface as synchronous - replication for streaming - replication. To do this, the streaming replication interface - (see ) must be used to stream out - data. Clients have to send Standby status update (F) - (see ) messages, just like streaming - replication clients do. - + + Overview + - - - A synchronous replica receiving changes via logical decoding will work in - the scope of a single database. Since, in contrast to - that, synchronous_standby_names currently is - server wide, this means this technique will not work properly if more - than one database is actively used. + + Logical decoding can be used to build + synchronous + replication solutions with the same user interface as synchronous + replication for streaming + replication. To do this, the streaming replication interface + (see ) must be used to stream out + data. Clients have to send Standby status update (F) + (see ) messages, just like streaming + replication clients do. + + + + + A synchronous replica receiving changes via logical decoding will work in + the scope of a single database. Since, in contrast to + that, synchronous_standby_names currently is + server wide, this means this technique will not work properly if more + than one database is actively used. + + + + + + Caveats + + Caveats + + + + The use of any command to take an ACCESS EXCLUSIVE lock on [user] catalog tables + can cause the deadlock of logical decoding in synchronous mode. This means that + at the transaction commit or prepared transaction, the command hangs or the server + becomes to block any new connections. To avoid this, users must refrain from such + operations. + + + + When COMMIT is conducted for a transaction that has + issued explicit LOCK on pg_class + with logical decoding, the deadlock occurs. Also, committing one that runs + CLUSTER pg_class is another + deadlock scenario. + + + + Similarly, executing PREPARE TRANSACTION + after LOCK command on pg_class and + logical decoding of published table within the same transaction leads to the deadlock. + Clustering pg_trigger by CLUSTER command + brings about the deadlock as well, when published table has a trigger and any operations + that will be decoded are conducted on the same table. + + + + The deadlock can happen when users execute TRUNCATE + on user_catalog_table under the condition that output plugin have reference to it. - + + @@ -1253,9 +1298,11 @@ stream_commit_cb(...); <-- commit of the streamed transaction The logical replication solution that builds distributed two phase commit using this feature can deadlock if the prepared transaction has locked - [user] catalog tables exclusively. They need to inform users to not have - locks on catalog tables (via explicit LOCK command) in - such transactions. + [user] catalog tables exclusively. To avoid this users must refrain from + having locks on catalog tables (e.g. explicit LOCK command) + in such transactions. + (See Synchronous + Replication Support for Logical Decoding for the details.)