Issues

Use replication lag as health metric

readysettech/readyset#1771 · opened

on Nov 6, 2026

Postgres: `SET "DateStyle" TO "ISO"`, `'iso'` or `'ISO, MDY'` silently turns the connection into a permanent proxy

readysettech/readyset#1770 · opened

on Nov 5, 2026

Snapshot spends about half its CPU on ICU sort keys for text when the database collation is C

readysettech/readyset#1769 · opened

on Nov 4, 2026

Postgres: SHOW CACHES, SHOW READYSET STATUS and EXPLAIN LAST STATEMENT fail over the extended protocol

readysettech/readyset#1768 · opened

on Nov 4, 2026

Postgres: Readyset leaves the socket open after Terminate, so asyncpg's close() never returns

readysettech/readyset#1767 · opened

on Nov 4, 2026

sslmode=verify-full in UPSTREAM_DB_URL stops Readyset with an unrelated "percent-encoded" error

readysettech/readyset#1766 · opened

on Nov 4, 2026

Postgres: a shallow cache never stores an empty result over the extended protocol

readysettech/readyset#1765 · opened

on Nov 4, 2026

Deep cache: GROUP BY on an expression fully materializes the aggregate over the whole table

readysettech/readyset#1764 · opened

on Nov 4, 2026

CREATE CACHE under a new name for a query with a deep cache leaves the query uncached once an open connection reads it again

readysettech/readyset#1763 · opened

on Nov 4, 2026

Postgres: a client connection opened while the upstream is down never gets an upstream, so it fails every statement that needs Postgres after Postgres is back

readysettech/readyset#1762 · opened

on Nov 4, 2026

Postgres: replicated UPDATEs apply about 20 times slower than INSERTs at the default REPLICATION_BATCH_SIZE, and writes to every other table wait behind them

readysettech/readyset#1761 · opened

on Nov 4, 2026

TIME(p) values are rendered with six fractional digits, or none, instead of p digits (text protocol), and a zero TIME can arrive as -00:00:00 (binary protocol)

readysettech/readyset#1760 · opened

on Oct 30, 2026

1

Query sampler reads the previous query's upstream result after one upstream timeout, and reports false mismatches until restart

readysettech/readyset#1759 · opened

on Oct 29, 2026

1

Postgres: proxying an unnamed extended-protocol statement adds about 0.4 ms, against under 0.1 ms for a named statement or a simple query

readysettech/readyset#1758 · opened

on Oct 28, 2026

MySQL deep cache: a non-numeric string in a logical OR (`name OR '!'`, `name || '!'`) evaluates as true and returns the string `'1'`; MySQL returns the integer 0

readysettech/readyset#1757 · opened

on Oct 28, 2026

Deep caches refuse `col LIKE $1` and `a = $1 AND b >= $2`, the form every server-side-binding ORM sends

readysettech/readyset#1756 · opened

on Oct 28, 2026

Postgres: support `FETCH FIRST n ROWS ONLY` and `OFFSET n ROWS`, which Hibernate uses for every paginated query

readysettech/readyset#1755 · opened

on Oct 28, 2026

MySQL: `SET character_set_results = NULL`, which Connector/J sends on every connection, makes Readyset proxy every query on that connection, so Java applications get no cache hits

readysettech/readyset#1754 · opened

on Oct 28, 2026

2

Postgres: once a deep cache exists, a psycopg 3 query whose integer parameter changes size fails with "invalid buffer size" and Readyset closes the connection

readysettech/readyset#1753 · opened

on Oct 28, 2026

1

MySQL deep cache: `LIKE BINARY 'prefix%'` returns no rows

readysettech/readyset#1752 · opened

on Oct 28, 2026

MySQL: Readyset doesn't advertise CLIENT_TRANSACTIONS, so mysqlclient sessions stay in autocommit and rollback() undoes nothing

readysettech/readyset#1751 · opened

on Oct 28, 2026

MySQL replicator retries a failed upstream connection every second with no backoff and logs two lines per attempt

readysettech/readyset#1750 · opened

on Oct 28, 2026

Replication lag poller leaves one MySQL connection behind per second while `SHOW MASTER STATUS` is blocked

readysettech/readyset#1749 · opened

on Oct 28, 2026

1

TRUNCATE of a replicated table blocks its domain for a time quadratic in the row count (250k rows: 71 s, 1M: 21 min, 4M: hours)

readysettech/readyset#1748 · opened

on Oct 28, 2026

1

Replication batch size (GROUP_COMMIT_MAX_TRX): an auto mode that adapts to the backlog, and changeable without a restart

readysettech/readyset#1747 · opened

on Oct 28, 2026

Navigation Menu

Repository navigation

Issues

Use replication lag as health metric

Postgres: `SET "DateStyle" TO "ISO"`, `'iso'` or `'ISO, MDY'` silently turns the connection into a permanent proxy

Snapshot spends about half its CPU on ICU sort keys for text when the database collation is C

Postgres: SHOW CACHES, SHOW READYSET STATUS and EXPLAIN LAST STATEMENT fail over the extended protocol

Postgres: Readyset leaves the socket open after Terminate, so asyncpg's close() never returns

sslmode=verify-full in UPSTREAM_DB_URL stops Readyset with an unrelated "percent-encoded" error

Postgres: a shallow cache never stores an empty result over the extended protocol

Deep cache: GROUP BY on an expression fully materializes the aggregate over the whole table

CREATE CACHE under a new name for a query with a deep cache leaves the query uncached once an open connection reads it again

Postgres: a client connection opened while the upstream is down never gets an upstream, so it fails every statement that needs Postgres after Postgres is back

Postgres: replicated UPDATEs apply about 20 times slower than INSERTs at the default REPLICATION_BATCH_SIZE, and writes to every other table wait behind them

TIME(p) values are rendered with six fractional digits, or none, instead of p digits (text protocol), and a zero TIME can arrive as -00:00:00 (binary protocol)

Query sampler reads the previous query's upstream result after one upstream timeout, and reports false mismatches until restart

Postgres: proxying an unnamed extended-protocol statement adds about 0.4 ms, against under 0.1 ms for a named statement or a simple query

MySQL deep cache: a non-numeric string in a logical OR (`name OR '!'`, `name || '!'`) evaluates as true and returns the string `'1'`; MySQL returns the integer 0

Deep caches refuse `col LIKE $1` and `a = $1 AND b >= $2`, the form every server-side-binding ORM sends

Postgres: support `FETCH FIRST n ROWS ONLY` and `OFFSET n ROWS`, which Hibernate uses for every paginated query

MySQL: `SET character_set_results = NULL`, which Connector/J sends on every connection, makes Readyset proxy every query on that connection, so Java applications get no cache hits

Postgres: once a deep cache exists, a psycopg 3 query whose integer parameter changes size fails with "invalid buffer size" and Readyset closes the connection

MySQL deep cache: `LIKE BINARY 'prefix%'` returns no rows

MySQL: Readyset doesn't advertise CLIENT_TRANSACTIONS, so mysqlclient sessions stay in autocommit and rollback() undoes nothing

MySQL replicator retries a failed upstream connection every second with no backoff and logs two lines per attempt

Replication lag poller leaves one MySQL connection behind per second while `SHOW MASTER STATUS` is blocked

TRUNCATE of a replicated table blocks its domain for a time quadratic in the row count (250k rows: 71 s, 1M: 21 min, 4M: hours)

Replication batch size (GROUP_COMMIT_MAX_TRX): an auto mode that adapts to the backlog, and changeable without a restart