PolarProxy release notes

Category

Description

New Features & Performance Optimizations

DynamoDB compatibility for PolarDB for PostgreSQL Distributed Edition clusters.

DynamoDB Scan / Query / ListTables now include a default limit of 100 items returned per response to prevent slow responses caused by full table scans.

DynamoDB supports the TagResource, UntagResource, and ListTagsOfResource tag management APIs.

DynamoDB newly supports strict validation mode. Currently, this mode returns a validation error when duplicate primary keys are detected within the same table during batch read operations.

Queries on the pg_stat_activity view now broadcast the now() function call to all nodes.

Bug Fixes

Fixed an issue where DynamoDB Scan / Query on non-existent tables or indexes could cause the proxy process to crash.

Fixed an issue where the DynamoDB BatchGetItem response contained an incorrect field name for unprocessed items.

Fixed a parsing error in the DynamoDB KeyConditionExpression BETWEEN operator; validation errors are now returned when the operand types on both sides are inconsistent.

Fixed a parsing error for attribute alias expressions such as a.#b in DynamoDB expressions.

Fixed two issues in PostgreSQL session parameter replay to read-only nodes: transient errors could interrupt replay, causing subsequent read-only query results to diverge from the primary node, and transaction_read_only was incorrectly replayed to read-only nodes.

Category

Description

New Features & Performance Optimizations

Optimized large-query routing for PostgreSQL to prevent proxy crashes on read-only endpoints.

DynamoDB now supports the returnItemCollectionMetrics parameter for write operation metrics.

Enabled backend connection pool reuse for DynamoDB to improve connection utilization.

Bug Fixes

Fixed DynamoDB transaction API (TransactWriteItems/TransactGetItems) to correctly return TransactionConflict errors during conflicts.

The proxy process no longer crashes when a PostgreSQL client startup packet is missing the user or database parameter.

Response packets are now correctly forwarded when flow control is triggered on a PostgreSQL backend connection.

Category

Description

Enhancements

Disabled proxy syslog by default.

Bug fixes

Fixed console API call timeouts during primary node failover.

Category

Description

New Features & Performance Optimizations

DynamoDB now supports global consistency.

DynamoDB now supports the UpdateTable command.

Improved the DynamoDB error message for unsupported kernel functions.

Bug Fixes

Fixed an issue where the proxy process might fail to start if the only read-only (RO) node becomes unavailable for a Read-only address whose Load Balancing Policy is set to Connections-based Load Balancing.

Fixed a potential connection leak in the transaction-level connection pool under specific scenarios.

Fixed a crash in the database proxy after cluster switchover when the always_to_master parameter is enabled.

Fixed case-sensitive HTTP header handling in the DynamoDB compatibility feature.

Fixed incorrect InternalServerError response in the DynamoDB compatibility feature when a TransactWriteItems conflict occurs at BEGIN ISOLATION LEVEL SERIALIZABLE.

Category

Description

New features and performance optimizations

DynamoDB now supports database version validation.

Fixes

The role of a read-only node with a columnar index is now correctly identified after a switchover.

Global consistency checks no longer time out when the primary node is unavailable.

Category

Description

New features and performance optimizations

Supports PolarDB for PostgreSQL Distributed Edition clusters.

DynamoDB compatibility now supports the DescribeTimeToLive and UpdateTimeToLive APIs.

Bug fixes

Fixed a response packet error that occurred when executing a query with the extended protocol on keep-alive connections.

Category

Description

New features & performance optimizations

Added support for the SCRAM-SHA-256 authentication mechanism.

Enhanced connection persistence for faster failover recovery during switchovers.

Fixed issues

Fixed misidentification of columnar index read-only node status.

Corrected the server/client encoding parameter returned during connection establishment.

Added support for the skipEmptyExchange connection parameter to the document database compatibility feature.

Corrected the error message and error code returned for connection failures.

Category

Description

New features & performance optimizations

Added MongoDB-compatible connection endpoints.

Added the FORCE_IMCI_NODES hint to route query requests to a columnar index read-only node.

Improved session parameter handling in the transaction-level connection pool.

Improved connection reuse for RO nodes in transaction-level connection pool mode.

Enhanced the internal monitoring and probing mechanism.

Bug fixes

Fixed inaccurate parameter values returned by the server during connection setup.

Fixed result set merge failures when broadcasting queries on the pg_stat_activity view.

Fixed a bug requiring two cancel requests to stop a running query in non-connection pool mode.

Fixed reduced transaction-level connection pool reuse rate caused by the application_name session parameter.

Fixed route forwarding failure caused by oversized Bind protocol messages from clients.

Fixed connection interruptions when calling the BatchGetItem API in the DynamoDB compatibility feature.

Category

Description

New Features and Performance Optimizations

Added DynamoDB API compatibility for a custom endpoint.

Improved performance with a large number of databases.

Bug Fixes

Fixed connection pool fill-up caused by held connection requests during node suspension.

Fixed improper replay of internal errors when functions like nextval fail, which caused session errors.

Fixed missing tables in connection pool mode due to non-strict search_path matching.

Fixed incorrect replay of subsequent session commands caused by a session parameter replay error.

Fixed incorrect routing of some UNION ALL queries to the primary node during read/write splitting.

Category

Description

New Features and Performance Optimizations

• Improved transaction-level connection pool reuse rate with the extended query protocol.

• Added support for persistent connections in the transaction-level connection pool.

Fixed Issues

• Fixed startup session parameters not taking effect in transaction-level connection pool mode.

• Fixed connections in a "Transaction Error" state entering the transaction-level connection pool and affecting other sessions.

• Fixed newly added SQL throttling rules not applying to existing connections.

• Fixed SQL throttling not taking effect after an unexpected proxy restart.

• Fixed incorrect routing of replayed statements that caused duplicate prepared statement errors when functions like nextval fail.

Category

Description

Bug fix

Fixed issues introduced by the on-demand replay of Parse/Prepare statements.

Category

Description

New features and performance optimizations

• Supports on-demand replay of Parse and Prepare statements, converting broadcast mode to unicast mode.

• Supports SQL throttling template validation.

• Supports dynamic configuration of send and receive buffer sizes.

Fixed issues

Fixed an issue where TLS 1.0 connections failed in a JDBC environment.

Category

Description

New features and performance optimizations

• Added support for SQL throttling.

• Added support for on-demand connections.

Fixed issues

• Fixed an issue where multi-statement queries were parsed incorrectly, causing routing errors.

• Fixed configuration invalidation when replaying list-type parameters such as search_path.

Category

Description

New Features & Performance Optimizations

Optimized short transaction forwarding.

Fixed Issues

• Fixed a node routing issue on read-only endpoints using active request-based load balancing during high latency.

• Fixed a failed Parse message replay causing connection closure.

• Fixed a potential memory leak from failed Prepare statement executions.

Category

Description

New Features and Performance Optimizations

The consistency level now supports global consistency.

Bug Fixes

• Fixed a database proxy crash caused by a stack overflow when the database returned a large Error Msg.

• Fixed a potential conflict when enabling or disabling the advanced transaction splitting feature for multiple custom endpoints.

• Fixed a parsing failure for mixed-protocol messages.

• Fixed a memory leak in the handling of Close messages.

Category

Description

New features and performance optimizations

• Supports SSL mutual authentication, forced SSL, Certificate Revocation List (CRL), and TLS 1.3.

• Supports message splitting.

• Supports serverless cluster APIs.

• Introduced master_failure_mode, which allows read-only nodes to serve requests if the primary node disconnects.

• Added the enable_master_retry_req parameter, which defaults to false, to stop read-only nodes from forwarding failed requests to the primary node.

• Optimized the replay logic to combine and send Parse and Close messages in a single operation.

• A connection to a read-only node can now be established if authentication succeeds on the primary node, even if it fails on the read-only node.

Fixed issues

• Fixed Parse messages marked as master_read not being sent to other nodes.

• Fixed unrecognized type_mask type in Prepare statements when releasing resources with Deallocate.

• Fixed incorrect is_superuser array length.

• Fixed incorrect error message returned on permission verification failure.

• Fixed a crash caused by enabling SSL without the ssl_version parameter.

• Fixed a crash caused by a failure to promptly terminate the processing of an abnormal auth packet.

Category

Description

New features and performance optimizations

• Adds support for persistent connections.

• Enhances read/write splitting routing for SQL statements that contain functions.

Fixed issues

• Fixed transaction status inconsistency on a read-only node after disconnect and reconnect.

• Fixed connections to read-only nodes not being returned to the pool in connection pool mode.

Category

Description

Fixed issues

• Fixed inconsistent transaction status on read-only endpoints.

• Fixed two memory leaks.

• Fixed an out-of-bounds issue with C-type strings when parsing startup messages.

Category

Description

New Feature & Performance Optimization

Read-only endpoint load balancing now supports active request count-based balancing.

Bug Fixes

• Fixed message discarding in transaction-level connection pool mode caused by a MySQL function misidentifying PostgreSQL message types.

• Fixed a kernel crash when calling the /v1/maxscale API to adjust the threads parameter in version 2.3.47.

Category

Description

Issue fixes

Fixed a connection reuse issue in the transaction-level connection pool that occurred when a statement failed within a transaction.

Category

Description

Enhancements

Added transaction-level connection pooling.

Bug Fixes

Fixed a read-only endpoint crash.

Category

Description

New feature and performance optimization

Added the /*force_session_to_master*/ hint to ensure all requests in a session are routed to the primary node.

Fixed issues

• Fixed unbalanced load when using the LRU algorithm.

• Fixed an issue where the /*force_master*/ hint failed to take effect when an SQL statement contained special characters like \t\n.

• Fixed a memory leak during PostgreSQL authentication.

• Fixed a crash on the read-only endpoint.

Category

Description

Fixed issues

• Fixed a load-balancing issue with read-only endpoints.

• Fixed a crash in clusters with more than 32 CPU cores.

Category

Description

Fixed issues

• The control plane is now compatible with PolarDB for PostgreSQL 14 when refreshing configurations.

• The database now validates the IP address and port of the database proxy when pgmon probes the database.

• Function names in the whitelist are now case-insensitive.

• Retrieving the whitelist count no longer causes the database proxy to continuously pull the whitelist.

Category

Description

New Features and Performance Optimizations

Added graceful restart for database instances.

Fixed Issues

Optimized transaction splitting. The transaction splitting switch on the proxy now controls the corresponding switch on the database.

Category

Description

Fixed issue

Configured TCP KeepAlive parameters in the database proxy instead of relying on system parameters, ensuring consistent behavior across environments.

Category

Description

Fixed issues

• Stopped broadcasting parse messages with write attributes to read-only nodes.

• Optimized the routing of global temporary tables.

• Optimized the management of table name and function name whitelists.

• Changed the unit for read-only delay to milliseconds, with a default of 300 ms.

Category

Description

Fixed issues

Read-only endpoints no longer check the message length.

Category

Description

fixed issues

• A read-only node now remains in the running state if a connection probe times out; other types of probe failures will still change its state.

• If a connection probe to a read-only node times out, the backend no longer routes new client connections to that node.

• Routing can now identify and route global temporary tables.

• Fixed memory bloat caused by numerous broadcast requests.

Category

Description

New features and performance optimizations

• The monitor now displays the values of internal database variables.

• Added an option to force routing to the primary database.

Fixed issues

• The monitor now initializes internal variables from the configuration upon startup.

• Added support for traffic throttling.

Category

Description

Fixed issues

• When splitting extended query messages, the last of any consecutive E/v messages now serves as the split boundary.

• By default, extended query messages are no longer split.

• Parse message replays now include parameter information.

• The monitor now establishes a single connection per probe cycle and uses it for all other probes within that cycle.

Category

Description

Fixed issues

• Optimized routing whitelists for functions and tables. Removed 'nextval' from the function whitelist and added system tables to the table whitelist.

• The routing node for Bind is now recorded. Execute messages are now routed separately using this recorded node.

• Extended Query is now split, using Bind as the start boundary and Execute as the end boundary.

• If a transaction routed to the primary database does not include an Execute message, the system treats it as having no write operations. This allows subsequent read requests to be routed to a read-only database.

• SYNC messages are no longer routed until all results for broadcast messages have been received.

• Fixed an issue where client connections would disconnect if a read-only node disconnected during connection establishment.

• Added the enable_split_ext_pkts switch. When this switch is off, the system does not split Extended Query message sequences.

• The system no longer routes requests after the primary database connection closes.

• Fixed an issue where new requests would stall when all read-only nodes were unavailable.

• The system now broadcasts transaction start, commit, and rollback statements.

Category

Description

New features & performance optimizations

Added support for routing read-after-write requests to read-only nodes within RC isolation level transactions.

Fixed issues

• Optimized read-only node reconnection. A disconnected read-only node no longer reconnects immediately. It now reconnects only after the next monitor probe confirms the node is healthy.

• Fixed the primary node not being auto-added when the read/write endpoint was stopped.

• Fixed "unhealthy database" errors returned to clients when a database connection failed.

Category

Description

bug fix

Fixed a dual-primary issue where the flag on the old primary node was removed due to a connection failure when a new primary node was added.

Category

Description

New features and performance optimizations

Added support for custom endpoints.

Fixed issues

• If a read/write cluster endpoint does not have a primary node, the monitor automatically adds the primary node to the endpoint.

• Connections no longer stall when a client sends a sync message after a proxy broadcast and before a read-only node returns a result.

Category

Description

Fixed issues

• For Extended Query message splitting, the system now extracts message sequences containing Begin or Set from repeated message sequences to process and send them separately.

• During transaction splitting, if a request is routed to a read-only node while a connection transaction is active, the system now updates the transaction status in the packet to in transaction.

Category

Description

New features and performance optimizations

PostgreSQL now supports transaction splitting.

Fixed issues

• The system now checks if the whitelist table exists before querying its content.

• The system now routes a multi-statement containing only SET and read requests directly to the primary node instead of broadcasting it.

Category

Description

New features & performance optimizations

• Read-only nodes now automatically reconnect after an unexpected connection loss.

• You can now add nodes dynamically.

Fixed issues

• Fixed an issue where deleting a read-only node via an API call failed to disconnect its corresponding read-only connection.

• Added support for edb protocol messages.

• The proxy now retrieves a password from the database, if needed, before establishing a connection.

Category

Description

New feature & performance optimization

Introduced a dynamic whitelist for function and table names.

Bug fix

Added JSON-related functions to the proxy's built-in function whitelist.

Category

Description

Fixed issues

• The switch allowing the primary node to accept read operations is now disabled by default.

• Fixed an issue where SSL was blocking the proxy's main thread.

Category

Description

New features & performance optimizations

• SSL support.

• Synchronous connection establishment.

• Request cancellation.

• Probing for database version information with monitor.

• Merging of multi-node results for the pg_stat_activity, pg_locks, and pg_stat_ssl views.

Issue fixes

Fixed a memory leak in the parser.

Category

Description

Fixed issues

• The execute statement now inherits the hint from the prepare statement.

• Fixed an issue where the SET operation did not lock the primary node in a stored procedure or transaction.

• The proxy no longer crashes when a query is made while the primary database is not accepting reads and all read replicas have a weight of zero.

• Allowing the primary database to accept reads no longer affects existing connections.

Category

Description

Fixed issues

• Enabled TCP keepalive after the TCP three-way handshake.

• Added specific read request functions to the function whitelist.

Category

Description

New features and performance optimizations

Added hint syntax (/* force_master */ and /* force_slave */).

Fixed issues

• Optimized multi-statement routing. The primary node is not locked if the statement contains only set and select.

• The system now replays the original request to the primary node when a read-only node returns an error.

• Fixed an issue where show statements were routed to the primary node.

Category

Description

Bug fixes

• Added support for the PostgreSQL protocol.

• The monitor can now probe the liveness status, primary/standby status, LSN offset, and replication delay (in microseconds) of PolarDB for PostgreSQL clusters.

• Added support for read/write splitting.

• Added support for session-level and table-level read consistency. Session-level read consistency is enabled by default, while table-level read consistency is disabled by default due to its limitations.