v1.5.1.1

Fixes

• After the server closes an idle pooled connection (a Postgres restart, pg_terminate_backend, a server-side idle timeout), the first session on it can fail with UnexpectedRowCountStatementError instead of a connection error. Hasql maps the extra result that reports the closed connection to that error, and a singleRow statement that returns two rows fails the same way on a healthy connection, so the error alone cannot justify a discard: the dead connection got returned to the pool and failed one more session before it was discarded, so after a database restart every pooled connection failed two sessions. Now, when a session fails with an error that does not require a discard, use also checks the status that the adapter reports for the connection and discards it when it is marked bad.

v1.5.1.0

Fixes

• Fix pool capacity leak when use gets interrupted during connection establishment, connection refresh, or an observation (#60)
• Fix pool state corruption from throwing observation handlers; handlers are now contained (see the updated observationHandler docs)
• Fix background reaper dying permanently from a failed connection release or observation
• Fix release leaking flushed entries on failure
• Fix waiters waiting forever on a garbage-collected pool

Added

• reuseOrder setting (#59): LIFO reuse lets the pool shrink to the connections the load needs and keeps caches warm; FIFO stays the default
• One background timer for acquisition timeouts instead of one uncancelable timer per waiting request (#58)

Changed

• Connections are handed to waiting requests in arrival order; a request no longer takes a connection while others wait (#58)

v1.5.0.1

• Allow newer pqi.

v1.5.0.0

Breaking

• Migrate to hasql-2, replacing postgresql-libpq with the pqi connection-adapter interface. acquire now takes a Pqi.Adapter as its first argument; pick one from an adapter package such as pqi-ffi or pqi-native.

v1.4.2.3

Fixes

• Fix publishing.

v1.4.2.2

Fixes

• Fix pool capacity leak when initSession fails (#56)
• Fix background reaper never passively evicting idle connections: it used the connection-lifetime timeout instead of the idleness timeout when checking for idle connections

v1.4.2.1

Fixes

• Discard pooled connections after driver errors (#55)

v1.4

• Migrated to hasql-1.10
• Updated connection settings API to use monoid-based Settings instead of list-based [Setting]
• Updated error types to use Hasql.Errors module instead of Hasql.Session and Hasql.Connection
• Changed session execution API from Session.run session connection to Connection.use connection session
• Error handling now uses ConnectionSessionError for connection-level issues instead of ClientError
• Updated statement construction to use Statement.preparable and Statement.unpreparable instead of direct constructor
• Hid the Defaults module from the public API

v1.3

• Adapt to the new settings model of hasql-1.9

v1.2

• Migrated to hasql-1.7
• Changed references to QueryError in observations to SessionError

v1.1

• ReadyForUseConnectionStatus got extended with the ConnectionReadyForUseReason details.
• initSession setting added.

v1

• Optional observability event stream added. Provides a flexible mechanism for monitoring the healthiness of the pool via logs and metrics.
• Configuration got isolated into a DSL, which will allow to provide new configurations without breaking backward compatibility.

v0.10.1

• Avoid releasing connections on exceptions thrown in session

v0.9

• Maximal lifetime added for connections. Allows to refresh the connections in time cleaning up the resources.

Breaking:

• The acquisition timeout is now non-optional.
• Moved to DiffTime for timeouts.

v0.8.0.7

Fix excessive connections during releases due to race conditions.

v0.8.0.5

Fix connections not returning to the pool on exceptions.

v0.8.0.2

Fixed Windows build.

v0.8

release became reusable. You can use it to destroy the whole pool (same as before), but now also you can use it to reset the connections.

Acquisition timeout added.

Breaking changes in API:

• Removed PoolIsReleasedUsageError
• acquire extended with the acquisition timeout parameter
• acquireDynamically extended with the acquisition timeout parameter

v0.7.2

Added support for dynamic connection configuration (issue #11).

v0.7.1.2

Fixed connections not being released if they were in use during the call to release.

v0.7.1

Added Exception for UsageError.

v0.7

Simplified the implementation a lot by removing the notion of timeout.

Breaking:

• Removed the Settings type
• Changed the signature of acquire

v0.6

Moved away from “resource-pool” and fixed the handling of lost connections.

Breaking:

• Changed the suffix of UsageError constructors from Error to UsageError
• Added PoolIsReleasedUsageError