Getting Started
DocumentDB v1.0-0 was released on 2026-10-19. It is built from the same commit as the second release candidate, v1.0-RC2. The changelog lists the engine changes.
| Channel | 1.0 version | Platforms |
|---|---|---|
| Linux packages | packages 1.0.0, extension 1.0-0 | Ubuntu 24.04 and RHEL-compatible 9, PostgreSQL 17 and 18, amd64 and arm64 |
| Container image | pg15-1.0.0 to pg18-1.0.0; latest is pg17-1.0.0 | Linux amd64 and arm64 |
| Kubernetes operator | operator 1.0.0, which runs engine 1.0-0 | Kubernetes 1.33+ with the ImageVolume feature gate on (on by default from 1.35, GA in 1.36) |
The SQL extension version is 1.0-0 everywhere, and hello reports it in internal.documentdb_versions.
documentdb-local image reports its version, platform and CPU architecture to Scarf at startup and hourly, on by default. Native packages send nothing. See Usage telemetry to turn it off.find and aggregate with batchSize: 0 now check that the caller can read the collection. $merge and $out now need insert or update permission on the target collection, not just read. createUser and updateUser reject passwords that fail SASLprep.count and the $stdDevPop, $stdDevSamp, $median and $percentile accumulators, and applies to the symbol type. Collation is still off by default; see Known issues.documentdb.enableRoleCrud on (off by default), createRole and dropRole keep custom roles in a roles catalog. updateRole is still unsupported.compact reports bytesFreed as the measured change in on-disk size.InternalError, and exits with an error instead of panicking when PostgreSQL is unreachable at startup.$getField/$setField/$unsetField on system variables, $firstN/$lastN without a sort, $documents inside $facet, $near/$geoNear ordering on 2dsphere indexes, overlapping dropIndexes deadlocks, and several RUM index scan and vacuum fixes.--init-data-path script now stops the container. A stale postmaster.pid is removed on the next start only when the volume's lock shows a 1.0 container wrote it; otherwise the container refuses to start. The container exits if PostgreSQL inside it stops. TOAST compression defaults to lz4, and --disable-extended-rum is deprecated and ignored. See Container lifecycle.default_toast_compression = 'lz4'; pass DOCUMENTDB_TOAST_COMPRESSION=pglz or default through sudo to keep another setting.mongodump and mongorestore. See Upgrading the image.The setup admin can't create or drop users or roles (#752). createUser, and createRole with documentdb.enableRoleCrud on, fail with User is not authorized to perform this action when run as the admin user created at setup or by the container. On a package install, manage users on the host instead:
printf 'Password for app: '; IFS= read -r -s APP_PASSWORD; echo
printf '%s' "$APP_PASSWORD" | sudo documentdb-gateway-admin create-user --username app \
--password-stdin --roles '[{"role":"readAnyDatabase","db":"admin"}]'
sudo documentdb-gateway-admin drop-user --username appdistinct or count on a grouped view can crash the server. distinct on a view whose pipeline ends in $group with an accumulator such as $sum, or count with a query on such a view, crashes the PostgreSQL backend. PostgreSQL restarts and disconnects every other session.
A newer container image doesn't upgrade an existing volume (#773). The image never runs ALTER EXTENSION ... UPDATE, and a changed --password on an existing volume is ignored.
A 1.0 container can refuse a volume an older container used. If the volume still has an older container's postmaster.pid, the 1.0 image reports that it exists and refuses to start. Stop the old container first: for 0.116 and 0.117, docker stop removes the file. Images 0.114 and earlier don't normally stop PostgreSQL cleanly and usually leave it behind. Once nothing else uses the volume, re-create the container with -e DOCUMENTDB_FORCE_REMOVE_STALE_POSTMASTER_PID=true, then re-create it again without the variable, because it skips the in-use check on every start.
A single-document update can miss a match. An update with multi: false, or a findAndModify, whose filter uses fields other than _id can report no match when PostgreSQL runs it with a generic plan; with upsert it then inserts a duplicate. This is rare with default settings; setting plan_cache_mode = force_custom_plan in the PostgreSQL configuration avoids it.
An expensive $replaceAll or $regexFindAll can outlive maxTimeMS. A very costly call on a large string can keep running after maxTimeMS expires and can't be cancelled.
A write can be applied twice (#767). If the backend connection is lost after a write commits but before the reply, the gateway retries it, so a non-idempotent update such as $inc can be applied twice even with retryWrites=false.
$lookup can fail on a collection with a wildcard index (#769), with Unrecognized BSON data type code 0.
Collation is off by default. find, aggregate, count and delete with a collation fail as unsupported unless documentdb_core.enableCollation is on. distinct, findAndModify and update don't support collation at all.
Container options that do nothing. --log-level and --enable-telemetry are validated but have no effect, and a --data-path that contains a space fails.
Open packaging and container issues:
--password argument is readable in the container's process list; pass -e PASSWORD instead.Open an issue in documentdb/documentdb with your DocumentDB and PostgreSQL versions, and either the package versions or the Release Version line from docker logs.