The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.

NAME

MongoDB::Upgrading - Deprecations and behavior changes from the v0 driver

VERSION

version v0.999.999.4

DESCRIPTION

The v1 driver represents a substantial step forward in functionality and consistency. There are many areas where the old API has been deprecated or changed in a backward breaking way.

This document is intended to help developers update their code to take into account API changes from the v0 driver to the v1 driver.

RATIONALE

Changes to the driver were deemed necessary to achieve certain goals:

  • consistency (intra-driver) – many parts of the v0 API were inconsistent, behaving differently from method to method; the v1 API minimizes developer surprises by improving consistency in return types and exception types.

  • consistency (inter-driver) — "next-generation" MongoDB drivers across all languages are converging on common APIs and common behaviors; this simplifies developer education and support, as cross-language examples will be similar.

  • encapsulation – too many low-level, internal operations were exposed as part of the API, which complicates maintenance work; the v1 API aims to minimize the "public surface" available to developers.

  • abstraction – many methods returned raw server documents for end-user code to inspect, which is brittle in the face of changes by the server over time; the v1 API uses objects to abstract the details behind standard method calls.

  • compatibility – some new features or changes in the MongoDB server, like the client authentication model, no longer fit the old driver design.

  • portability – the v0 driver had a large dependency tree and required a compiler and various libraries; the v1 driver removes some dependencies and uses widely-used CPAN modules in place of custom C code where possible.

  • round-trippable data – the v0 BSON implementation could easily lead to data corruption when round-tripping documents; the v1 driver makes round-trippability a top priority, even if this means removing or changing how BSON encoding or decoding works

INSTALLATION AND DEPENDENCY CHANGES

SSL and SASL

The v0 driver required a compiler and OpenSSL and libgsasl for SSL and SASL support, respectively. The v1 driver instead relies on CPAN modules IO::Socket::SSL and Authen::SASL for SSL and SASL support, respectively.

SSL configuration is now possible via the ssl attribute.

Authentication configuration is described in "AUTHENTICATION" in MongoDB::MongoClient.

BEHAVIOR CHANGES

MongoClient configuration

New configuration options

Several configuration options have been added, with particular emphasis on adding more granular control of timings and timeout behaviors.

  • auth_mechanism

  • auth_mechanism_properties

  • bson_codec

  • connect_timeout_ms

  • heartbeat_frequency_ms

  • local_threshold_ms

  • max_time_ms

  • replica_set_name

  • read_pref_mode

  • read_pref_tag_sets

  • server_selection_timeout_ms

  • socket_check_interval_ms

  • socket_timeout_ms

Replica set configuration

Connecting to a replica set now requires a replica set name, given either with the replica_set_name option for MongoDB::MongoClient or with the replicaSet option in a connection string. For example:

    $client = MongoDB::MongoClient->new(
        host => "mongodb://rs1.example.com,rs2.example.com/",
        replica_set_name => 'the_set',
    );

    $client = MongoDB::MongoClient->new(
        host => "mongodb://rs1.example.com,rs2.example.com/?replicaSet=the_set"
    );

Configuration options changed to read-only

Configuration options are changing to be immutable to prevent surprising action-at-a-distance. (E.g. changing an attribute value in some part of the code changes it for other parts of the code that didn't expect it.) Going forward, options may be set at MongoDB::MongoClient construction time only.

The following options have changed to be read-only:

  • db_name

  • j

  • password

  • ssl

  • username

  • w

  • wtimeout

Write concern may be overridden at the MongoDB::Database and MongoDB::Collection level during construction of those objects. For more details, see the later section on write concern changes.

Mapping between connection string and configuration options

Many configuration options may be set via a connection string URI in the host option. In the v0 driver, the precedence between the connection string and constructor options was completely inconsistent. In the v1 driver, options set via a connection string URI will take precedence over options passed to the constructor. This is consistent with with other MongoDB drivers (as well as how DBI treats Data Source Names).

The list of servers and ports as well as the optional username, password and db_name options come directly from URI structure. Other options are parsed as key-value parameters at the end of the connection string. The following table shows how connection string keys map to configuration options in the MongoDB::MongoClient:

    Connection String Key           MongoClient option
    ---------------------------     -----------------------------
    authMechanism                   auth_mechanism
    authMechanismProperties         auth_mechanism_properties
    connectTimeoutMS                connect_timeout_ms
    heartbeatFrequencyMS            heartbeat_frequency_ms
    journal                         j
    localThresholdMS                local_threshold_ms
    maxTimeMS                       max_time_ms
    readPreference                  read_pref_mode
    readPreferenceTags              read_pref_tag_sets
    replicaSet                      replica_set_name
    serverSelectionTimeoutMS        server_selection_timeout_ms
    socketCheckIntervalMS           socket_check_interval_ms
    socketTimeoutMS                 socket_timeout_ms
    ssl                             ssl
    w                               w
    wTimeoutMS                      wtimeout

The readPreferenceTags and authMechanismProperties keys take colon-delimited, comma-separated pairs:

    readPreferenceTags=dc:nyeast,rack:1
    authMechanismProperties=SERVICE_NAME:mongodb

The readPreferenceTags option may be repeated to build up a list of tag set documents:

    readPreferenceTags=dc:nyc,rack:1&readPreferenceTags=dc:nyc

Deprecated configuration options

Several options have been superseded, replaced or renamed for clarity and are thus deprecated and undocumented. They are kept for a limited degree of backwards compatibility. They will be generally be used as fallbacks for other options. If any were read-write, they have also been changed to read-only.

  • dt_type — see "BSON encoding changes" for details.

  • query_timeout — replaced by socket_timeout_ms; if set, this will be used as a fallback default for socket_timeout_ms.

  • sasl — superseded by auth_mechanism; if set, this will be used along with sasl_mechanism as a fallback default for auth_mechanism.

  • sasl_mechanism — superseded by auth_mechanism; if set, this will be used as a fallback default for auth_mechanism.

  • timeout — replaced by connect_timeout_ms; if set, this will be used as a fallback default for connect_timeout_ms.

These will be removed in a future major release.

Configuration options removed

Some configuration options have been removed entirely, as they no longer serve any purpose given changes to server discovery, server selection and connection handling:

  • auto_connect

  • auto_reconnect

  • find_master

  • max_bson_size

As described further below in the "BSON encoding changes" section, these BSON encoding configuration options have been removed as well:

  • inflate_dbrefs

  • inflate_regexps

Removed configuration options will be ignored if passed to the MongoDB::MongoClient constructor.

Lazy connections and reconnections on demand

The improved approach to server monitoring and selection allows all connections to be lazy. When the client is constructed, no connections are made until the first network operation is needed. At that time, the client will scan all servers in the seed list and begin regular monitoring. Connections that drop will be re-established when needed.

See SERVER SELECTION and SERVER MONITORING AND FAILOVER in MongoDB::MongoClient for details.

Exceptions are the preferred error handling approach

In the v0 driver, errors could be indicated in various ways:

  • boolean return value

  • string return value is an error; hash ref is success

  • document that might contain an 'err', 'errmsg' or '$err' field

  • thrown string exception

Regardless of the documented error handling, every method that involved a network operation would throw an exception on various network errors.

In the v1 driver, exceptions objects are the standard way of indicating errors. The exception hierarchy is described in MongoDB::Error.

Cursors and query responses

In v0, MongoDB::Cursor objects were used for ordinary queries as well as the query-like commands aggregation and parallel scan. However, only cursor iteration commands worked for aggregation and parallel scan "cursors"; the rest of the MongoDB::Cursor API didn't apply and was fatal.

In v1, all result iteration is done via the new MongoDB::QueryResult class. MongoDB::Cursor is now just a thin wrapper that holds query parameters, instantiates a MongoDB::QueryResult on demand, and passes iteration methods through to the query result object.

This significantly simplifies the code base and should have little end-user visibility unless users are specifically checking the return type of queries and query-like methods.

The explain cursor method no longer resets the cursor.

The slave_ok cursor method now sets the read_preference to 'secondaryPreferred' or clears it to 'primary'.

The snapshot cursor method now requires a boolean argument, allowing it to be turned on or off before executing the query. Calling it without an argument (as it was in v0) is a fatal exception.

Parallel scan "cursors" are now QueryResult objects, with the same iteration methods as in v0.

The $MongoDB::Cursor::slave_ok global variable has been removed as part of the revision to read preference handling. See the "Read preference objects and the read_preference method" in read preferences section below for more details.

The $MongoDB::Cursor::timeout global variable has also been removed. Timeouts are set during MongoDB::MongoClient configuration and are immutable. See the section on "MongoClient configuration" in configuration changes for more.

Aggregation API

On MongoDB 2.6 or later, aggregate always uses a cursor to execute the query. The batchSize option has been added (but has no effect prior to 2.6). The cursor option is deprecated.

The return types for the aggregate method are now always QueryResult objects, regardless of whether the aggregation uses a cursor internally or is an 'explain'.

NOTE: To help users with a 2.6 mongos and mixed version shards with versions before 2.6, passing the deprecated 'cursor' option with a false value will disable the use of a cursor. This workaround is provided for convenience and will be removed when 2.4 is no longer supported.

Read preference objects and the read_preference method

A new MongoDB::ReadPreference class is used to encapsulate read preference attributes. In the v1 driver, it is constructed from the read_pref_mode and read_pref_tag_sets attributes on MongoDB::MongoClient:

    MongoDB::MongoClient->new(
        read_pref_mode => 'primaryPreferred',
        read_pref_tag_sets => [ { dc => 'useast' }, {} ],
    );

The old read_preference method to change the read preference has been removed and trying to set a read preference after the client has been created is a fatal error.

The read_preference method now returns the MongoDB::ReadPreference object generated from read_pref_mode and read_pref_tag_sets.

It is inherited by MongoDB::Database, MongoDB::Collection, and MongoDB::GridFS objects unless provided as an option to the relevant factory methods:

    my $coll = $db->get_collection(
        "foo", { read_preference => 'secondary' }
    );

Such read_preference arguments may be a MongoDB::ReadPreference object, a hash reference of arguments to construct one, or a string that represents the read preference mode.

MongoDB::Database and MongoDB::Collection also have clone methods that allow easy alteration of a read preference for a limited scope.

    my $coll2 = $coll->clone( read_preference => 'secondaryPreferred' );

For MongoDB::Cursor, the read_preference method sets a hidden read preference attribute that is used for the query in place of the MongoDB::MongoClient default read_preference attribute. This means that calling read_preference on a cursor object no longer changes the read preference globally on the client – the read preference change is scoped to the cursor object only.

Write concern objects and removing the safe argument

A new MongoDB::WriteConcern class is used to encapsulate write concern attributes. In the v1 driver, it is constructed from the w, wtimeout and j attributes on MongoDB::MongoClient:

    MongoDB::MongoClient->new( w => 'majority', wtimeout => 1000 );

The write_concern method now returns the MongoDB::WriteConcern object generated from w, wtimeout and j.

It is inherited by MongoDB::Database, MongoDB::Collection, and MongoDB::GridFS objects unless provided as an option to the relevant factory methods:

    $db = $client->get_database(
        "test", { write_concern => { w => 'majority' } }
    );

Such write_concern arguments may be a MongoDB::WriteConcern object, a hash reference of arguments to construct one, or a string that represents the w mode.

MongoDB::Database and MongoDB::Collection also have clone methods that allow easy alteration of a write concern for a limited scope.

    my $coll2 = $coll->clone( write_concern => { w => 1 } );

The safe argument is no longer used in the new CRUD API.

Authentication based only on configuration options

Authentication now happens automatically on connection during the "handshake" with any given server based on the auth_mechanism attribute.

The old authenticate method in MongoDB::MongoClient has been removed.

Bulk API

Bulk method names changed to match CRUD API

Method names match the new CRUD API, e.g. insert_one instead of insert and so one. The legacy names are deprecated.

Bulk insertion

Insertion via the bulk API will NOT insert an _id into the original document if one does not exist. Previous documentation was not specific whether this was the case or if the _id was added to the document sent to the server.

Bulk write results

The bulk write results class has been renamed to MongoDB::BulkWriteResult. It keeps MongoDB::WriteResult as an empty superclass for some backwards compatibility so that $result->isa("MongoDB::WriteResult") will continue to work as expected.

The attributes have been renamed to be consistent with the new CRUD API. The legacy names are deprecated, but are available as aliases.

GridFS

The MongoDB::GridFS class now has explicit read preference and write concern attributes inherited from MongoDB::MongoClient or MongoDB::Database, just like MongoDB::Collection. This means that GridFS operations now default to an acknowledged write concern, just like collection operations have been doing since v0.502.0 in 2012.

The use of safe is deprecated.

Support for ancient, undocumented positional parameters circa 2010 has been removed.

Low-level functions removed

Low-level driver functions have been removed from the public API.

MongoDB::Connection removed

The MongoDB::Connection module was deprecated in v0.502.0 and has been removed.

BSON encoding changes

In the v1 driver, BSON encoding and decoding have been encapsulated into a MongoDB::BSON codec object. This can be provided at any level, from MongoDB::MongoClient to MongoDB::Collection. If not provided, a default will be created that behaves similarly to the v0 encoding/decoding functions, except for the following changes.

$MongoDB::BSON::use_binary removed

Historically, this defaulted to false, which corrupts binary data when round tripping. Retrieving a binary data element and re-inserting it would have resulted in a field with UTF-8 encoded string of binary data.

Going forward, binary data will be returned as a MongoDB::BSON::Binary object. A future driver may add the ability to control decoding to allow alternative representations.

$MongoDB::BSON::use_boolean removed

This global variable never worked. BSON booleans were always deserialized as boolean objects. A future driver may add the ability to control boolean representation.

$MongoDB::BSON::utf8_flag_on removed

In order to ensure round-tripping of string data, this variable is removed. BSON strings will always be decoded to Perl character strings. Anything else risks double-encoding a round-trip.

$MongoDB::BSON::looks_like_number and $MongoDB::BSON::char deprecated and re-scoped

In order to allow a future driver to provide more flexible user-customized encoding and decoding, these global variables are deprecated. If set, they will be examined during MongoDB::MongoClient->new() to set the configuration of a default MongoDB::BSON codec (if one is not provided). Changing them later will NOT change the behavior of the codec object.

MongoDB::MongoClient option inflate_regexps removed

Previously, BSON regular expressions decoded to qr{} references by default and the MongoDB::MongoClient inflate_regexps option was available to decode instead to MongoDB::BSON::Regexps.

Going forward in the v1.0.0 driver, for safety and consistency with other drivers, BSON regular expressions always decode to MongoDB::BSON::Regexp objects.

MongoDB::MongoClient option inflate_dbrefs removed

The inflate_dbrefs configuration option has been removed and replaced with a dbref_callback option in MongoDB::BSON.

By default, the MongoDB::MongoClient will create a MongoDB::BSON codec that will construct MongoDB::DBRef objects. This ensures that DBRefs properly round-trip.

MongoDB::MongoClient option dt_type deprecated and changed to read-only

The dt_type option is now only takes effect if MongoDB::MongoClient constructs a MongoDB::BSON codec object. It has been changed to a read-only attribute so that any code that relied on changing dt_type after constructing a MongoDB::MongoClient object will fail instead of being silently ignored.

Int32 vs Int64 encoding changes

On 64-bit Perls, integers that fit in 32-bits will be encoded as BSON Int32 (whereas previously these were always encoded as BSON Int64).

Math::BigInt objects will always be encoded as BSON Int64, which allows users to force 64-bit encoding if desired.

Added support for Time::Moment

Time::Moment is a much faster replacement for the venerable DateTime module. The BSON codec will serialize Time::Moment objects correctly and can use that module as an argument for the dt_type codec attribute.

Added support for encoding common JSON boolean classes

Most JSON libraries on CPAN implement their own boolean classes. The following libraries boolean types will now encode correctly as BSON booleans:

  • JSON::XS

  • Cpanel::JSON::XS

  • JSON::PP

  • JSON::Tiny

  • Mojo::JSON

DBRef objects

The fetch method and related attributes client, verify_db, and verify_coll have been removed from MongoDB::DBRef.

Providing a fetch method was inconsistent with other MongoDB drivers, which either never provided it, or have dropped it in the next-generation drivers. It requires a client attribute, which tightly couples BSON decoding to the client model, causing circular reference issues and triggering Perl memory bugs under threads. Therefore, the v1.0.0 driver no longer support fetching directly from MongoDB::DBRef; users will need to implement their own methods for dereferencing.

Additonally, the db attribute is now optional, consistent with the specification for DBRefs.

Also, all attributes (ref, id and db) are now read-only, consistent with the move toward immutable objects throughout the driver.

To support round-tripping DBRefs with additional fields other than $ref, $id and $db, the DBRef class now has an attribute called extra. As not all drivers support this feature, using it for new DBRefs is not recommended.

DEPRECATED METHODS

Deprecated options and methods may be removed in a future release. Their documentation has been removed to discourage ongoing use. Unless otherwise stated, they will continue to behave as they previously did, allowing a degree of backwards compatibility until code is updated to the new MongoDB driver API.

MongoDB::Database

  • last_error — Errors are now indicated via exceptions at the time database commands are executed.

MongoDB::Collection

  • insert, batch_insert, remove, update, save, query and find_and_modify — A new common driver CRUD API replaces these legacy methods.

  • get_collection — This method implied that collections could be contained inside collection. This doesn't actually happen so it's confusing to have a Collection be a factory for collections. Users who want nested namespaces should be explicit and create them off Database objects instead.

  • ensure_index, drop_indexes, drop_index, get_index — A new MongoDB::IndexView class is accessable through the indexes method, offering greater consistency in behavior across drivers.

  • validate — The return values have changed over different server versions, so this method is risky to use; it has more use as a one-off tool, which can be accomplished via run_command.

MongoDB::CommandResult

  • result — has been renamed to 'output' for clarity

MongoDB::Cursor

  • slave_ok — this modifier method is superseded by the 'read_preference' modifier method

  • count — this is superseded by the "MongoDB::Collection#count" in MongoDB::Collection count method. Previously, this ignored skip/limit unless a true argument was passed, which was a bizarre, non-intuitive and inconsistent API.

MongoDB::BulkWrite and MongoDB::BulkWriteView

  • insert — renamed to 'insert_one' for consistency with CRUD API

  • update — renamed to 'update_many' for consistency with CRUD API

  • remove — renamed to 'delete_many' for consistency with CRUD API

  • remove_one — renamed to 'delete_one' for consistency with CRUD API

AUTHORS

  • David Golden <david@mongodb.com>

  • Mike Friedman <friedo@friedo.com>

  • Kristina Chodorow <k.chodorow@gmail.com>

  • Florian Ragwitz <rafl@debian.org>

COPYRIGHT AND LICENSE

This software is Copyright (c) 2015 by MongoDB, Inc..

This is free software, licensed under:

  The Apache License, Version 2.0, January 2004