Incompatible changes in ArangoDB 3.1

It is recommended to check the following list of incompatible changes before upgrading to ArangoDB 3.1, and adjust any client programs if necessary.

Communication Layer

The internal commication layer is now based on Boost ASIO. A few options regarding threads and communication have been changed.

There are no longer two different threads pools (--scheduler.threads and --server.threads). The option --scheduler.threads has been removed. The number of threads is now controlled by the option --server.threads only. By default --server.threads is set to the number of hyper-cores.

As a consequence of the change, the following (hidden) startup options have been removed:

  • --server.extra-threads
  • --server.aql-threads
  • --server.backend
  • --server.thread-affinity


The behavior of the AQL array comparison operators has changed for empty arrays:

  • ALL and ANY now always return false when the left-hand operand is an empty array. The behavior for non-empty arrays does not change:
    • [] ALL == 1 will return false
    • [1] ALL == 1 will return true
    • [1, 2] ALL == 1 will return false
    • [2, 2] ALL == 1 will return false
    • [] ANY == 1 will return false
    • [1] ANY == 1 will return true
    • [1, 2] ANY == 1 will return true
    • [2, 2] ANY == 1 will return false
  • NONE now always returns true when the left-hand operand is an empty array. The behavior for non-empty arrays does not change:
    • [] NONE == 1 will return true
    • [1] NONE == 1 will return false
    • [1, 2] NONE == 1 will return false
    • [2, 2] NONE == 1 will return true
  • WITH in cluster traversals is now mandatory in order to avoid deadlocks.

Data format changes

The attribute maximalSize has been renamed to journalSize in collection meta-data files (“parameter.json”). Files containing the maximalSize attribute will still be picked up correctly for not-yet adjusted collections.

The format of the revision values stored in the _rev attribute of documents has been changed in 3.1. Up to 3.0 they were strings containing largish decimal numbers. With 3.1, revision values are still strings, but are actually encoded time stamps of the creation date of the revision of the document. The time stamps are acquired using a hybrid logical clock (HLC) on the DB-Server that holds the revision (for the concept of a hybrid logical clock see this paper). See this manual section for details.

ArangoDB >= 3.1 can ArangoDB 3.0 database directories and will simply continue to use the old _rev attribute values. New revisions will be written with the new time stamps.

It is highly recommended to backup all your data before loading a database directory that was written by ArangoDB <= 3.0 into an ArangoDB >= 3.1.

To change all your old _rev attributes into new style time stamps you have to use arangodump to dump all data out (using ArangoDB 3.0), and use arangorestore into the new ArangoDB 3.1, which is the safest way to upgrade.

The change also affects the return format of _rev values and other revision values in HTTP APIs (see below).

HTTP API changes

APIs added

The following HTTP REST APIs have been added for online log level adjustment of the server:

  • GET /_admin/log/level returns the current log level settings
  • PUT /_admin/log/level modifies the current log level settings

APIs changed

  • the following REST APIs that return revision ids now make use of the new revision id format introduced in 3.1. All revision ids returned will be strings as in 3.0, but have a different internal format.

    The following APIs are affected:

    • GET /_api/collection/{collection}/checksum: revision attribute
    • GET /_api/collection/{collection}/revision: revision attribute
    • all other APIs that return documents, which may include the documents’ _rev attribute

    Client applications should not try to interpret the internals of revision values, but only use revision values for checking whether two revision strings are identical.

  • the replication REST APIs will now use the attribute name journalSize instead of maximalSize when returning information about collections.

  • the default value for keepNull has been changed from false to true for the following partial update operations for vertices and edges in /_api/gharial:

    • PATCH /_api/gharial/{graph}/vertex/{collection}/{key}
    • PATCH /_api/gharial/{graph}/edge/{collection}/{key}

    The value for keepNull can still be set explicitly to false by setting the URL parameter keepNull to a value of false.

  • the REST API for dropping collections (DELETE /_api/collection) now accepts an optional query string parameter isSystem, which can set to true in order to drop system collections. If the parameter is not set or not set to true, the REST API will refuse to drop system collections. In previous versions of ArangoDB, the isSystem parameter did not exist, and there was no distinction between system and non-system collections when dropping collections.

  • the REST API for retrieving AQL query results (POST /_api/cursor) will now return an additional sub-attribute loading collections that will contain the total time required for loading and locking collections during the AQL query when profiling is enabled. The attribute can be found in the extra result attribute in sub-attribute loading collections. The attribute will only be set if profiling was enabled for the query.

  • the REST API for retrieving AQL query results (POST /_api/cursor) will now accept the optional attribute memoryLimit.

Foxx Testing

The QUnit interface to Mocha has been removed. This affects the behavior of the suite, test, before, after, beforeEach and afterEach functions in Foxx test suites. The suite and test functions are now provided by the TDD interface. The before, after, beforeEach and afterEach functions are now provided by the BDD interface.

This should not cause any problems with existing tests but may result in failures in test cases that previously passed for the wrong reasons. Specifically the execution order of the before, after, etc functions now follows the intended order and is no longer arbitrary.

For details on the expected behavior of these functions see the testing chapter in the Foxx documentation.