Server parameters

All parameters are prefixed pg_xpatch. and take effect only when pg-xpatch is loaded through shared_preload_libraries. They are defined in pg_xpatch.c.

Change context

restart (postmaster): set in postgresql.conf or with ALTER SYSTEM, then restart the server.
superuser (superuser): a superuser can change it at runtime with ALTER SYSTEM + SELECT pg_reload_conf().
session (user): any user can SET it per session.

Where a range shows ≥ N, the practical ceiling is available memory.

Content cache

The big one: reconstructed content, shared across all backends.

Parameter	Default	Range	Context	Controls
`pg_xpatch.cache_size_mb`	256 MB	≥ 1	restart	total shared memory for the content cache
`pg_xpatch.cache_max_entries`	65536	≥ 1000	restart	how many entries the cache can hold
`pg_xpatch.cache_max_entry_kb`	256 KB	≥ 16	superuser	entries larger than this are skipped, not cached
`pg_xpatch.cache_slot_size_kb`	4 KB	1 to 64	restart	content storage granularity inside the cache
`pg_xpatch.cache_partitions`	32	1 to 256	restart	lock stripes; higher allows more concurrent backends

cache_max_entry_kb is the only content-cache setting you can change without a restart. See Tuning read performance.

Lookup caches

Smaller shared caches that keep sequence bookkeeping off the hot path.

Parameter	Default	Range	Context	Controls
`pg_xpatch.group_cache_size_mb`	16 MB	≥ 1	restart	group to max-`_xp_seq` cache (INSERT)
`pg_xpatch.tid_cache_size_mb`	16 MB	≥ 1	restart	TID to seq cache (READ)
`pg_xpatch.seq_tid_cache_size_mb`	16 MB	≥ 1	restart	seq to TID reverse cache (reconstruction)

Inserts and encoding

Parameter	Default	Range	Context	Controls
`pg_xpatch.insert_cache_slots`	16	≥ 1	restart	how many (table, group) pairs can hold an active insert FIFO
`pg_xpatch.max_delta_columns`	32	≥ 1	restart	maximum delta-compressed columns per table
`pg_xpatch.encode_threads`	0	0 to 64	session	worker threads for parallel delta encoding (`0` = sequential)

encode_threads parallelizes the candidate-base encodings that compress_depth produces, so it only helps when compress_depth is above 1 and the content is large enough to be worth the coordination. At the default compress_depth of 1 it does nothing.

Cache warming

Parameter	Default	Range	Context	Controls
`pg_xpatch.warm_cache_workers`	4	≥ 0	session	default background workers for `warm_cache_parallel()` (`0` = sequential)

The effective worker count is also capped by PostgreSQL's max_worker_processes.

A starting point

A reasonable postgresql.conf block for a read-heavy deployment with room to grow:

shared_preload_libraries = 'pg_xpatch'pg_xpatch.cache_size_mb     = 1024   # fit the hot working setpg_xpatch.cache_max_entry_kb = 1024  # if you store larger valuespg_xpatch.warm_cache_workers = 8     # faster bulk warming

Tuning read performance

How to size these from cache_stats() readings.

Caching & performance

What each cache actually does.