Zero 1.4

Performance and Reliability Improvements

Installation

Copynpm install @rocicorp/zero@1.4

Upgrading

userID: "anon"

If you are using userID: 'anon' for logged-out users, change it to userID: null or userID: undefined. Passing userID: 'anon' is being deprecated in advance of upcoming security work in Zero 1.5.

Change this:

Copyimport {ZeroProvider} from '@rocicorp/zero/react'
 
return (
  <ZeroProvider userID={userID ?? 'anon'}>
    <App />
  </ZeroProvider>
)

To this:

Copyimport {ZeroProvider} from '@rocicorp/zero/react'
 
return (
  <ZeroProvider userID={userID}>
    <App />
  </ZeroProvider>
)

Features

• Kysely Adapter: Added zeroKysely, so Zero server mutators can run against a Kysely transaction.
• Analyze Query Library: Added @rocicorp/zero/analyze, for easier CLI-based query analysis.
• Shadow Sync: zero-cache can now periodically test initial sync in the background, to catch schema drift and full-resync failures earlier.
• Client Log Sink: Added a logSink param to Zero constructor for redirecting Zero client logs.
• Cloud Zero Ad 🙃: Added a small notice about Cloud Zero in the zero-cache-dev startup output. Disable with ZERO_ENABLE_STARTUP_MESSAGE=0.

Performance

• 7x faster client-side index construction
• 2x faster backfills by switching zero-cache to binary COPY
• Various optimizations to query hydration and updates:
  • +9-16% hydration and +3-18% updates by reducing polymorphism
  • +8-14% updates by reducing allocations
  • +5-16% hydration by reducing allocations
  • +7% hydration for queries with joins
• Avoid full table scans from reordered compound primary keys

Fixes

• Pagination on non-null columns could trigger full scans (thanks @utkarsh-611!)
• Include details in query/mutator validation errors (thanks @tjenkinson!)
• Inspector query analysis wasn't using Zero's planner
• Add query name to hydration spans in otel
• PgBouncer causing cached plan must not change result type error
• Transient litestream restore failures were incorrectly alerting
• Nullable array columns silently converted to [] (thanks @kvnkusch!)
• Change-log entries could be incorrectly purged during handoff
• Concurrent, nested, and CREATE/DROP INDEX CONCURRENTLY DDL were breaking schema replication
• Event-trigger upgrade permission errors sometimes swallowed
• Replication slot creation hanging during startup
• Autoreset during startup could deadlock while recreating a replication slot
• Original SQLite transaction failures hidden behind rollback errors
• Columns with complex Postgres backfilled incorrectly
• Latency histograms buckets are too broad
• Inspector parse and analyze errors incorrectly emitting ERROR logs
• Stalled change-streamer transactions could hang progress indefinitely
• OTel metrics from multiple worker processes clobbering each other
• Drain-mode SIGTERM exited uncleanly during startup

Breaking Changes

None. One API was deprecated:

• userID: 'anon' for logged-out clients

See Upgrading for more information.