SynapCores Community Edition v2.4.4-ce
Community Edition · free · seamless upgrade from v2.4.x
v2.4.4-ce makes concurrent writes behave the way developers coming from PostgreSQL, SQLite,
or MySQL/InnoDB expect — no false conflicts, real SERIALIZABLE and write-skew-safe
REPEATABLE READ, and schema-heavy workloads that stay fast no matter how long they run.
If you've been building high-concurrency OLTP or agent workloads on SynapCores, this is the
release that removes the asterisks. A database's job under concurrency is to conflict only
when two transactions genuinely collide — and never otherwise. v2.4.4-ce gets you there.
Concurrent inserts into constrained tables no longer falsely conflict
Two clients inserting different rows into the same table — even one with a UNIQUE column
or a FOREIGN KEY — now both commit without a retry. Only a genuine duplicate value or a
real row-level collision produces a conflict. This completes the concurrent-write
correctness work begun in v2.4.3: v2.4.3 fixed the cross-table (shared-parent) case, and
v2.4.4 fixes the same-table UNIQUE/FK case. On a high-concurrency agent write workload,
engine-side conflict failures dropped from 21 to 3 — and the remaining 3 are test-harness
timing, not the engine. If you load-tested write-heavy, constraint-heavy tables before and
saw conflict/retry churn under contention, re-run that test.
Standard isolation levels, drop-in
SERIALIZABLE is now supported, and REPEATABLE READ is write-skew-safe — the same
guarantees you get from SQLite and InnoDB's repeatable read. Read-only transactions keep a
stable snapshot, so backups and long reads never abort, and a writer aborts only when
something it actually read was changed underneath it. Blind inserts register no reads and
never conflict.
Schema-heavy workloads stay fast over time
Long-running services that create and drop tables continuously now keep flat latency.
CREATE TABLE that used to drift from ~10 ms toward 4 seconds over a long create/drop cycle
now stays fast, thanks to automatic background storage maintenance that keeps churny catalog
stores compacted off the request path.
Multi-tenant schema changes don't stall each other
A schema change on one database no longer forces retries for unrelated statements working
against a different database. Admission is scoped per database, so one tenant's DDL is
invisible to another's writes.
Two SQL-surface fixes
ORDER BY an aggregate's output alias now sorts correctly against the projected result, and
SynapCores accepts SQLite's SELECT RAISE(ABORT|FAIL|ROLLBACK|IGNORE, 'msg') trigger-guard
idiom — so more off-the-shelf SQLite schemas run unchanged.
Upgrade
Drop-in. The on-disk format is backward-compatible — no migration, no export/reload. Pull
the new image or binary and restart.
docker pull synapcores/community:latest # v2.4.4-ce
Linux (x86_64 / aarch64) and macOS (Apple Silicon) binaries are on the
releases page.
Community Edition, free, live now on :latest across GHCR and Docker Hub.