Module Description

This module implements XEP-0313 (Message Archive Management). It enables a service to store all user messages for one-to-one chats as well as group chats (MUC, MultiUser Chat). It uses XEP-0059: Result Set Management for paging. It is a highly customizable module, that requires some skill and knowledge to operate properly and efficiently.

MongooseIM is compatible with MAM 0.4-0.6.

Configure MAM with different storage backends:

  • RDBMS (databases like MySQL, PostgreSQL, MS SQL Server)
  • Riak KV (NOSQL)
  • Cassandra (NOSQL)
  • ElasticSearch (NOSQL)

mod_mam_meta is a meta-module that ensures all relevant mod_mam_* modules are loaded and properly configured.

Message retraction

This module supports XEP-0424: Message Retraction with RDBMS storage backends. When a retraction message is received, the MAM module finds the message to retract and replaces it with a tombstone. The following criteria are used to find the original message:

  • The id attribute specified in the apply-to element of the retraction message has to be the same as the id attribute of the origin-id element of the original message.
  • Both messages need to originate from the same user.
  • Both messages need to be addressed to the same user.

If more than one message matches the criteria, only the most recent one is retracted. To avoid this case, it is recommended to use a unique identifier (UUID) as the origin ID.

This module allows message filtering by their text body (if enabled, see Common backend options). This means that an XMPP client, while requesting messages from the archive may not only specify standard form fields (with, start, end), but also full-text-search (of type text-single). If this happens, the client will receive only messages that contain words specified in the request.

The exact behaviour, like whether word ordering matters, may depend on the storage backend in use. For now rdbms backend has very limited support for this feature, while cassandra does not support it at all. riak and elasticsearch backends, on the other hand, should provide you with the best results when it comes to text filtering.

mod_mam_rdbms_arch returns all messages that contain all search words, order of words does not matter. Messages are sorted by timestamp (not by relevance).

Note on full text search with ElasticSearch backend

When using ElasticSearch MAM backend, the value provided in full-text-search form field will be passed to ElasticSearch as Simple Search Query. If you're using our official ElasticSearch mappings from priv/elasticsearch then the query analyzer is set to english. Also note that the default separator for the search query is AND (which roughly means that ElasticSearch will search for messages containing all the terms provided in the query string).


  • backend (atom, default: rdbms) - Database backend to use. rdbms, riak, cassandra and elasticsearch are supported.
  • no_stanzaid_element (boolean, default: false) - Do not add a <stanza-id/> element from MAM v0.6.
  • is_archivable_message (module, default: mod_mam_utils) - Name of a module implementing is_archivable_message/3 callback that determines if the message should be archived. Warning: if you are using MUC Light, make sure this option is set to the MUC Light domain.
  • archive_chat_markers (boolean, default: false) - If set to true, XEP-0333 chat markers will be archived. See more details here
  • pm (list | false, default: []) - Override options for archivization of one-to-one messages. If the value of this option is false, one-to-one message archive is disabled.
  • muc (list | false, default: false) - Override options for archivization of group chat messages. If the value of this option is false, group chat message archive is disabled.
  • extra_lookup_params (atom, default: undefined) - a module implementing mam_iq behaviour. If this option has value other then undefined, function extra_lookup_params/2 from this module will be called when building MAM lookup parameters. This can be used to extend currently supported MAM query fields by a custom field or fields. This field(s) can be added to lookup params later passed to MAM backend.
  • message_retraction (boolean, default: true) - Enables XEP-0424: Message Retraction. This functionality is currently implemented only for the rdbms backend. Retraction messages are always archived regardless of this option.

backend, no_stanzaid_element, is_archivable_message and message_retraction will be applied to both pm and muc (if they are enabled), unless overriden explicitly (see example below).

PM-specific options

  • archive_groupchats (boolean, default: true) - When enabled, MAM will store groupchat messages in recipients' individual archives. USE WITH CAUTION! May increase archive size significantly. Disabling this option for existing installation will neither remove such messages from MAM storage, nor will filter out them from search results. MongooseIM will print a warning on startup if pm MAM is enabled without archive_groupchats being explicitly set to a specific value. In one of the future MongooseIM releases this option will default to false (as it's more common use case and less DB-consuming) and the warning message will be removed.

MUC-specific options

  • host (string, default: "conference.@HOST@") - MUC host that will be archived if MUC archiving is enabled.


The example below presents how to override common option for muc module specifically.

{mod_mam_meta, [
  {backend, rdbms},
  {async_writer, true}, %% this option enables async writer for RDBMS backend
  {muc, [
    {async_writer, false} %% disable async writer for MUC archive only

RDBMS backend options

These options will only have effect when the rdbms backend is used:

  • cache_users (boolean, default: true) - Enables Archive ID to integer mappings cache.
  • rdbms_message_format (atom, default: internal) - When set to simple, stores messages in XML and full JIDs. When set to internal, stores messages and JIDs in internal format. Warning: Archive MUST be empty to change this option.
  • async_writer (boolean, default: true) - Enables an asynchronous writer that is faster than the synchronous one but harder to debug. The async writers store batches of messages with a certain delay (see flush_interval), so the results of the lookup operations executed right after message routing may be incomplete until the configured time passes.
  • flush_interval (integer, default: 2000) How often (in milliseconds) the buffered messages are flushed to a DB.
  • max_batch_size (integer, default, 30) Max size of the batch insert query for an async writer. If the buffer is full, messages are flushed to a database immediately and the flush timer is reset.

Common backend options

  • user_prefs_store (atom, default: false) - Leaving this option as false will prevent users from setting their archiving preferences. It will also increase performance. Other possible values are:
    • rdbms (RDBMS backend only) - User archiving preferences saved in RDBMS. Slow and not recommended, but might be used for simplicity (keeping everything in RDBMS).
    • cassandra (Cassandra backend only) - User archiving preferences are saved in Cassandra.
    • mnesia (recommended) - User archiving preferences saved in Mnesia and accessed without transactions. Recommended in most deployments, could be overloaded with lots of users updating their preferences at once. There's a small risk of an inconsistent (in a rather harmless way) state of the preferences table.
  • full_text_search (boolean, default: true) - Enables full text search in message archive (see Full Text Search paragraph). Please note that the full text search is currently only implemented for rdbms and riak backends. Also, full text search works only for messages archived while this option is enabled.

is_archivable_message/3 callback

is_archivable_message option has to name a module exporting is_archivable_message/3 function conforming to the spec:

-spec is_archivable_message(Mod :: module(), Dir :: incoming | outgoing,
                          Packet :: exml:element()) -> boolean().

Servers SHOULD NOT archive messages that do not have a <body/> child tag. Servers SHOULD NOT archive delayed messages.

By default, all messages that hold meaningful content, rather than state changes such as Chat State Notifications, are archived.

Archiving chat markers

Archiving chat markers can be enabled by setting archive_chat_markers option to true. However it only works if is_archivable_message callback module is set to mod_mam_utils or isn't set at all.

When performing full text search chat markers are treated as if they had empty message body.

Riak backend

The Riak KV backend for MAM stores messages in weekly buckets so it's easier to remove old buckets. Archive querying is done using Riak KV 2.0 search mechanism called Yokozuna. Your instance of Riak KV must be configured with Yokozuna enabled.

This backend works with Riak KV 2.0 and above, but we recommend version 2.1.1.

Riak-specific options
  • bucket_type (default <<"mam_yz">>) - Riak bucket type.

  • search_index (default <<"mam">>) - Riak index name.

Cassandra backend

Please consult Outgoing connections page to learn how to properly configure Cassandra connection pool. By default, mod_mam Cassandra backend requires global pool with default tag:

{outgoing_pools, [
    {cassandra, global, default, [], []}.

ElasticSearch backend

First, make sure that your ElasticSearch cluster has expected indexes and mappings in place. Please consult Outgoing connections page to learn how to properly configure ElasticSearch connection pool.

Example configuration

{mod_mam_meta, [
        {backend, rdbms},

        {no_stanzaid_element, true},

        {pm, [{user_prefs_store, rdbms}]},
        {muc, [
               {host, ""},
               {rdbms_message_format, simple},
               {async_writer, false},
               {user_prefs_store, mnesia}


If you'd like to learn more about metrics in MongooseIM, please visit MongooseIM metrics page.

Name Type Description (when it gets incremented)
[Host, modMamArchiveRemoved] spiral User's entire archive is removed.
[Host, modMamArchived] spiral A message is stored in user's archive.
[Host, modMamDropped] spiral A message couldn't be enqueued due to an overloaded async worker.
[Host, modMamDropped2] spiral A message couldn't be stored in the DB (and got dropped).
[Host, modMamDroppedIQ] spiral MAM IQ has been dropped due to: high query frequency/invalid syntax or type.
[Host, modMamFlushed] spiral Message was stored in a DB asynchronously.
[Host, modMamForwarded] spiral A message is sent to a client as a part of a MAM query result.
[Host, modMamLookups] spiral A MAM lookup is performed.
[Host, modMamSinglePurges] spiral A single purge request is processed by MAM.
[Host, modMamMultiplePurges] spiral A bulk purge request is processed by MAM.
[Host, modMamPrefsGets] spiral Archiving preferences have been requested by a client.
[Host, modMamPrefsSets] spiral Archiving preferences have been updated by a client.
[Host, modMucMamArchiveRemoved] spiral Room's entire archive is removed.
[Host, modMucMamArchived] spiral A message is stored in room's archive.
[Host, modMucMamForwarded] spiral A message is sent to a client as a part of a MAM query result from MUC room.
[Host, modMucMamLookups] spiral A MAM lookup in MUC room is performed.
[Host, modMucMamSinglePurges] spiral A single purge request for MUC room is processed by MAM.
[Host, modMucMamMultiplePurges] spiral A bulk purge request for MUC room is processed by MAM.
[Host, modMucMamPrefsGets] spiral MUC archiving preferences have been requested by a client.
[Host, modMucMamPrefsSets] spiral MUC archiving preferences have been updated by a client.
[Host, mod_mam_rdbms_async_pool_writer, per_message_flush_time] histogram Average time per message insert measured in an async MAM worker.
[Host, mod_mam_rdbms_async_pool_writer, flush_time] histogram Average time per flush of all buffered messages measured in an async MAM worker.
[Host, mod_mam_muc_rdbms_async_pool_writer, per_message_flush_time] histogram Average time per message insert measured in an async MUC MAM worker.
[Host, mod_mam_muc_rdbms_async_pool_writer, flush_time] histogram Average time per flush of all buffered messages measured in an async MUC MAM worker.
Backend action Description (when it gets incremented)
lookup A lookup in an archive.
archive One message is saved in an archive.