Hemibrain Release

Dvid executables for initial Hemibrain data release.

• Enforces no writing to metadata when server run in -readonly mode. This is necessary for running grayscale servers against Google-owned read-only GCS buckets.
• Deprecates single mutex for gbucket driver. GCS backend should be used for ingest-style storage and not multi-DVID parallel writes unless they are idempotent. (No races should be present on ingest for GCS storage.)
• Better separation of CORS and authorization option.

Maintenance Release

Changes since v0.9.3:

• add support for optional Pid-File generation
• improve error message for bad annotation block detection
• significant work to improve migration (export) features, including limiting the exported versions and writing to new databases without restarting the server.
• add POST /blocks to annotation
• update groupcache support for recent interface
• if no cors domains specified allow all origins if no authorization configured
• add admin token querystring to allow selective full write
• Pin ristretto to v0.0.1

Authentication and Simple Authorization

This release is among the last in the "legacy" branch that should honor the storage engines added before v1.0. Authentication through flyem-services is supported. After logging into flyem-services, you can call the /api/server/token to get a JWT that is optionally required for other HTTP requests via the Authorization header using the Bearer schema (see http://jwt.io/introduction). If the [auth] section in the configuration TOML is not set, no authentication and authorization is provided.

Bug Fix Release

Fixes a typo in the response from /api/repo/info.

Maintenance Release

Edit: Please do not use.

This release contains a typo in the /api/repos/info response. Fixed in v0.9.2.

---

• add mutation ids to annotation kafka msgs; fixes #308
• fix mutation id on merge op for annotations
• add get/set Tags to keyvalue, labelmap, annotation; fixes #335

Stable 0.9 Release

This is a stable release of dvid that can be used for current installations. A significant rewrite of dvid (v1.0), not guaranteed to be fully backward compatible with legacy data types and all storage engines initially, will be released by January. Pre-1.0 releases will work with current datastores and will not perform any auto-migration or metadata updates.

Significant changes since last release:

• Can use a branch head identifier ("uuid-in-repo" + ":" + "branch name") instead of a version UUID.
• Many fixes (e.g., repo deletion, race conditions)
• Performance improvements based on debugging under large-scale use at Janelia.
• Added ability for further customizations via TOML config (kafka queue size)
• Added Badger storage engine support
• Build system changes

Maintenance Release

• if kafka messages are too large, store in blobstore
• add transfer-data command to move kv pairs to new store
• add limit-versions to adjust metadata to only versions in a JSON file
• make delete optional for instance migration
• add HTTP API access to annotation reload

Maintenance Release

General enhancements:

• make sure get/put label index goes through cache layer
• don't send non-existant blocks for /specificblocks
• add mutation order id and data UUID to mutation log
• make data name parsing more robust
• add labelsz /counts with enhanced testing; fixes #306
• refactor TOML config, esp. for testing; add server-wide blobstore
• add GET /api/repo/<uuid>/branch-versions/<name>
• add ability to turn off downres for /split-supervoxel in labelmap
• add POST /maxlabel and complete POST /nextlabel; fixes #307
• add bulk GET /indices for labelmap; fixes #309
• load in-memory labelmap of master branch on startup

Logging changes:

• add kafka logging for POST /nextlabel and add documentation
• add optional user id to annotation mutation kafka messages; fixes #305
• add general POST logging and separate blobstore config

Bug fixes:

• fix uuid parsing
• prevent GC of underlying byte slices when using aliasing

For developers:

• Build scripts and guide are compatible with conda 4.6
• conda-recipe: Use conda compilers on both Mac and Linux

Maintenance Release

• allow setting root UUID in POST /repos
• require BlockSize for labelmap to be multiple of 16
• fix tag deletion on POST /elements
• deprecated GET /commit in favor of GET /status; fixes #295
• misc. fixes for annotations operations
• allow configurable shutdown delay; resolves #298
• support activity logging of # blocks for GET /specificblocks
• Fix a crash related to unsafe slice aliasing
• misc. logging improvements and fixes
• misc. stability, race condition fixes
• conda-recipe:
  • Upgrade to go-1.11
  • On Linux, use Ananconda's gcc compiler
  • build script enhancements

Maintenance Release

Note: The release binaries above include a new version of the dvid web console, which requires a slight tweak to your .toml file. Please add the following line to the [server] section:

[server]
webDefaultFile = "index.html"

Changes:

• Minimum mutation ID on startup can be specified in the TOML
• Allow deletion of key classes and versions
• Permit mirroring of all data instances
• Added flatten-mutations command
• Logging improvements
• Kafka enhancements:
  • Kafka activity topic is exposed in /api/server/info
  • Kafka topic suffixes are configurable for each data-instance
  • Allow kafka topic prefix configuration for mutation logging
  • Add annotation sync data to kafka logging
• Enhancements to /split-supervoxel
  • Clients can manually specify split/remain labels for /split-supervoxel (dangerous: power-users only)
  • Permit "empty" and "complete" supervoxel splits (previously, these were considered errors)
• Performance enhancements:
  • Point-queries (i.e. /labels) on labelmap instances can read directly from the compressed blocks (big speedup, especially for annotation syncs)
  • Reduce impact of /mappings request on concurrent mutations