path: root/lib/mnesia/doc/src/Mnesia_chap5.xmlsrc

<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">

<chapter>
  <header>
    <copyright>
      <year>1997</year><year>2013</year>
      <holder>Ericsson AB. All Rights Reserved.</holder>
    </copyright>
    <legalnotice>
      Licensed under the Apache License, Version 2.0 (the "License");
      you may not use this file except in compliance with the License.
      You may obtain a copy of the License at
 
          http://www.apache.org/licenses/LICENSE-2.0

      Unless required by applicable law or agreed to in writing, software
      distributed under the License is distributed on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
      See the License for the specific language governing permissions and
      limitations under the License.

    </legalnotice>

    <title>Miscellaneous Mnesia Features</title>
    <prepared>Claes Wikstr&ouml;m, Hans Nilsson and H&aring;kan Mattsson</prepared>
    <responsible></responsible>
    <docno></docno>
    <approved></approved>
    <checked></checked>
    <date></date>
    <rev></rev>
    <file>Mnesia_chap5.xml</file>
  </header>

  <p>The previous sections describe how to get started
    with <c>Mnesia</c> and how to build a <c>Mnesia</c> database. This
    section describes the more advanced features available
    when building a distributed, fault-tolerant <c>Mnesia</c> database.
    The following topics are included:</p>
  <list type="bulleted">
    <item>Indexing</item>
    <item>Distribution and fault tolerance</item>
    <item>Table fragmentation</item>
    <item>Local content tables</item>
    <item>Disc-less nodes</item>
    <item>More about schema management</item>
    <item><c>Mnesia</c> event handling</item>
    <item>Debugging <c>Mnesia</c> applications</item>
    <item>Concurrent processes in <c>Mnesia</c></item>
    <item>Prototyping</item>
    <item>Object-based programming with <c>Mnesia</c></item>
  </list>

  <section>
    <marker id="indexing"></marker>
    <title>Indexing</title>
    <p>Data retrieval and matching can be performed efficiently
      if the key for the record is known. Conversely, if the key is
      unknown, all records in a table must be searched. The larger the
      table, the more time consuming it becomes. To remedy this
      problem, <c>Mnesia</c> indexing capabilities are used to improve
      data retrieval and matching of records.</p>
    <p>The following two functions manipulate indexes on existing
      tables:</p>
    <list type="bulleted">
      <item><seealso marker="mnesia#add_table_index/2">mnesia:add_table_index(Tab, AttributeName)
       -> {aborted, R} |{atomic, ok}</seealso></item>
      <item><seealso marker="mnesia#del_table_index/2">mnesia:del_table_index(Tab, AttributeName)
       -> {aborted, R} |{atomic, ok}</seealso></item>
    </list>
    <p>These functions create or delete a table index on a field
      defined by <c>AttributeName</c>. To illustrate this, add an
      index to the table definition <c>(employee, {emp_no, name,
      salary, sex, phone, room_no})</c>, which is the example table
      from the <c>Company</c> database. The function that
      adds an index on element <c>salary</c> can be expressed
      as <c>mnesia:add_table_index(employee, salary)</c>.</p>
    <p>The indexing capabilities of <c>Mnesia</c> are used with the
      following three functions, which retrieve and match records
      based on index entries in the database:</p>
    <list type="bulleted">
      <item>
      <seealso marker="mnesia#index_read/3">mnesia:index_read(Tab, SecondaryKey, AttributeName)
       -> transaction abort | RecordList</seealso>
       avoids an exhaustive search of the entire table, by looking up
       <c>SecondaryKey</c> in the index to find the primary keys.
      </item>
      <item>
      <seealso marker="mnesia#index_match_object/2">mnesia:index_match_object(Pattern, AttributeName)
       -> transaction abort | RecordList</seealso>
       avoids an exhaustive search of the entire table, by looking up
       the secondary key in the index to find the primary keys.
       The secondary key is found in field <c>AttributeName</c> of
       <c>Pattern</c>. The secondary key must be bound.
      </item>
      <item>
      <seealso marker="mnesia#match_object/1">mnesia:match_object(Pattern)
       -> transaction abort | RecordList</seealso>
       uses indexes to avoid exhaustive search of the entire table.
       Unlike the previous functions, this function can use
       any index as long as the secondary key is bound.</item>
    </list>
    <p>These functions are further described and exemplified in
      <seealso marker="Mnesia_chap4#matching">Pattern Matching</seealso>.
    </p>
  </section>

  <section>
    <title>Distribution and Fault Tolerance</title>
    <p><c>Mnesia</c> is a distributed, fault-tolerant DBMS. Tables
      can be replicated on different Erlang nodes in various
      ways. The <c>Mnesia</c> programmer does not need to state
      where the different tables reside, only the names of the
      different tables need to be specified in the program code. This
      is known as "location transparency" and is an important
      concept. In particular:</p>
    <list type="bulleted">
      <item><p>A program works regardless of the data
       location. It makes no difference whether the data
       resides on the local node or on a remote node.</p>
       <p>Notice that the program runs slower if the data
         is located on a remote node.</p>
      </item>
      <item>The database can be reconfigured, and tables can be
       moved between nodes. These operations do not affect the user
       programs. 
      </item>
    </list>
    <p>It has previously been shown that each table has a number of
      system attributes, such as <c>index</c> and <c>type</c>.</p>
    <p>Table attributes are specified when the table is created. For
      example, the following function creates a table with two
      RAM replicas:</p>
    <pre>
      mnesia:create_table(foo,
                          [{ram_copies, [N1, N2]},
                           {attributes, record_info(fields, foo)}]).</pre>
    <p>Tables can also have the following properties,
      where each attribute has a list of Erlang nodes as its value:</p>
    <list type="bulleted">
      <item>
        <p><c>ram_copies</c>. The value of the node list is a list
          of Erlang nodes, and a RAM replica of the table resides on
          each node in the list.</p>
        <p>Notice that no disc operations are performed when
          a program executes write operations to these replicas.
          However, if permanent RAM replicas are required, the
          following alternatives are available:</p>
        <list type="ordered">
          <item>The function
          <seealso marker="mnesia#dump_tables/1">mnesia:dump_tables/1</seealso>
           can be used to dump RAM table replicas to disc.
          </item>
          <item>The table replicas can be backed up, either from
           RAM, or from disc if dumped there with this function.
          </item>
        </list>
      </item>
      <item><c>disc_copies</c>. The value of the attribute is a list
       of Erlang nodes, and a replica of the table resides both
       in RAM and on disc on each node in the list. Write operations
       addressed to the table address both the RAM and the disc
       copy of the table. 
      </item>
      <item><c>disc_only_copies</c>. The value of the attribute is a
       list of Erlang nodes, and a replica of the table resides
       only as a disc copy on each node in the list. The major
       disadvantage of this type of table replica is the access
       speed. The major advantage is that the table does not occupy
       space in memory.
      </item>
    </list>
    <p>In addition, table properties can be set and changed.
      For details, see
      <seealso marker="Mnesia_chap3#def_schema">Define a Schema</seealso>.
    </p>
    <p>There are basically two reasons for using more than one table
      replica: fault tolerance and speed. Notice
      that table replication provides a solution to both of these
      system requirements.</p>
    <p>If there are two active table replicas, all information is
      still available if one replica fails. This can be an
      important property in many applications. Furthermore, if a table
      replica exists at two specific nodes, applications that execute
      at either of these nodes can read data from the table without
      accessing the network. Network operations are considerably
      slower and consume more resources than local operations.</p>
    <p>It can be advantageous to create table replicas for a
      distributed application that reads data often, but writes data
      seldom, to achieve fast read operations on the local
      node. The major disadvantage with replication is the increased
      time to write data. If a table has two replicas, every write
      operation must access both table replicas. Since one of these
      write operations must be a network operation, it is considerably
      more expensive to perform a write operation to a replicated
      table than to a non-replicated table.</p>
  </section>

  <section>
    <title>Table Fragmentation</title>

    <section>
      <title>Concept</title>
      <p>A concept of table fragmentation has been introduced
        to cope with large tables. The idea is to split a
        table into several manageable fragments. Each fragment is
        implemented as a first class <c>Mnesia</c> table and can be
        replicated, have indexes, and so on, as any other table. But
        the tables cannot have <c>local_content</c> or have the
        <c>snmp</c> connection activated.</p>
      <p>To be able to access a record in a fragmented
        table, <c>Mnesia</c> must determine to which fragment the
        actual record belongs. This is done by module
        <c>mnesia_frag</c>, which implements the <c>mnesia_access</c>
        callback behavior. It is recommended to read the
        documentation about the function
        <seealso marker="mnesia#activity/4">mnesia:activity/4</seealso>
        to see how <c>mnesia_frag</c>
        can be used as a <c>mnesia_access</c> callback module.</p>
      <p>At each record access, <c>mnesia_frag</c> first computes
        a hash value from the record key. Second, the name of the
        table fragment is determined from the hash value.
        Finally the actual table access is performed by the same
        functions as for non-fragmented tables. When the key is
        not known beforehand, all fragments are searched for
        matching records.</p>
      <p>Notice that in <c>ordered_set</c> tables, the records
        are ordered per fragment, and the the order is undefined in
        results returned by <c>select</c> and <c>match_object</c>.</p>
      <p>The following code illustrates how a <c>Mnesia</c> table is
        converted to be a fragmented table and how more fragments
        are added later:</p>
      <code type="none"><![CDATA[
Eshell V4.7.3.3  (abort with ^G)
(a@sam)1> mnesia:start().
ok
(a@sam)2> mnesia:system_info(running_db_nodes).
[b@sam,c@sam,a@sam]
(a@sam)3> Tab = dictionary.
dictionary
(a@sam)4> mnesia:create_table(Tab, [{ram_copies, [a@sam, b@sam]}]).
{atomic,ok}
(a@sam)5> Write = fun(Keys) -> [mnesia:write({Tab,K,-K}) || K <- Keys], ok end.
#Fun<erl_eval>
(a@sam)6> mnesia:activity(sync_dirty, Write, [lists:seq(1, 256)], mnesia_frag).
ok
(a@sam)7> mnesia:change_table_frag(Tab, {activate, []}).
{atomic,ok}
(a@sam)8> mnesia:table_info(Tab, frag_properties).
[{base_table,dictionary},
 {foreign_key,undefined},
 {n_doubles,0},
 {n_fragments,1},
 {next_n_to_split,1},
 {node_pool,[a@sam,b@sam,c@sam]}]
(a@sam)9> Info = fun(Item) -> mnesia:table_info(Tab, Item) end.
#Fun<erl_eval>
(a@sam)10> Dist = mnesia:activity(sync_dirty, Info, [frag_dist], mnesia_frag).
[{c@sam,0},{a@sam,1},{b@sam,1}]
(a@sam)11> mnesia:change_table_frag(Tab, {add_frag, Dist}).
{atomic,ok}
(a@sam)12> Dist2 = mnesia:activity(sync_dirty, Info, [frag_dist], mnesia_frag).
[{b@sam,1},{c@sam,1},{a@sam,2}]
(a@sam)13> mnesia:change_table_frag(Tab, {add_frag, Dist2}).
{atomic,ok}
(a@sam)14> Dist3 = mnesia:activity(sync_dirty, Info, [frag_dist], mnesia_frag).
[{a@sam,2},{b@sam,2},{c@sam,2}]
(a@sam)15> mnesia:change_table_frag(Tab, {add_frag, Dist3}).
{atomic,ok}
(a@sam)16> Read = fun(Key) -> mnesia:read({Tab, Key}) end.
#Fun<erl_eval>
(a@sam)17> mnesia:activity(transaction, Read, [12], mnesia_frag).
[{dictionary,12,-12}]
(a@sam)18> mnesia:activity(sync_dirty, Info, [frag_size], mnesia_frag).
[{dictionary,64},
 {dictionary_frag2,64},
 {dictionary_frag3,64},
 {dictionary_frag4,64}]
(a@sam)19> 
      ]]></code>
    </section>

    <section>
      <title>Fragmentation Properties</title>
      <p>The table property <c>frag_properties</c> can be read with
        the function
        <seealso marker="mnesia#table_info/2">mnesia:table_info(Tab, frag_properties)</seealso>.
        The fragmentation properties are a list of tagged tuples with
        arity 2. By default the list is empty, but when it is
        non-empty it triggers <c>Mnesia</c> to regard the table as
        fragmented. The fragmentation properties are as follows:</p>
      <taglist>
        <tag><c>{n_fragments, Int}</c></tag>
        <item>
          <p><c>n_fragments</c> regulates how many fragments
            that the table currently has. This property can explicitly
            be set at table creation and later be changed with
            <c>{add_frag, NodesOrDist}</c> or
            <c>del_frag</c>. <c>n_fragments</c> defaults to <c>1</c>.</p>
        </item>
        <tag><c>{node_pool, List}</c></tag>
        <item>
          <p>The node pool contains a list of nodes and can
            explicitly be set at table creation and later be changed
            with <c>{add_node, Node}</c> or <c>{del_node, Node}</c>.
            At table creation <c>Mnesia</c> tries to distribute
            the replicas of each fragment evenly over all the nodes in
            the node pool. Hopefully all nodes end up with the
            same number of replicas. <c>node_pool</c> defaults to the
            return value from the function
            <seealso marker="mnesia#system_info/1">mnesia:system_info(db_nodes)</seealso>.</p>
        </item>
        <tag><c>{n_ram_copies, Int}</c></tag>
        <item>
          <p>Regulates how many <c>ram_copies</c> replicas
            that each fragment is to have. This property can
            explicitly be set at table creation. Defaults is
            <c>0</c>, but if <c>n_disc_copies</c> and
            <c>n_disc_only_copies</c> also are <c>0</c>,
            <c>n_ram_copies</c> defaults to <c>1</c>.</p>
        </item>
        <tag><c>{n_disc_copies, Int}</c></tag>
        <item>
          <p>Regulates how many <c>disc_copies</c> replicas that
            each fragment is to have. This property can explicitly
            be set at table creation. Default is <c>0</c>.</p>
        </item>
        <tag><c>{n_disc_only_copies, Int}</c></tag>
        <item>
          <p>Regulates how many <c>disc_only_copies</c> replicas
            that each fragment is to have. This property can
            explicitly be set at table creation. Defaults is
            <c>0</c>.</p>
        </item>
        <tag><c>{foreign_key, ForeignKey}</c></tag>
        <item>
          <p><c>ForeignKey</c> can either be the atom
            <c>undefined</c> or the tuple <c>{ForeignTab, Attr}</c>,
            where <c>Attr</c> denotes an attribute that is to be
            interpreted as a key in another fragmented table named
            <c>ForeignTab</c>. <c>Mnesia</c> ensures that the number of
            fragments in this table and in the foreign table are
            always the same.</p>
         <p>When fragments are added or deleted, <c>Mnesia</c>
            automatically propagates the operation to all
            fragmented tables that have a foreign key referring to this
            table. Instead of using the record key to determine which
            fragment to access, the value of field  <c>Attr</c> is
            used. This feature makes it possible to colocate records
            automatically in different tables to the same node.
            <c>foreign_key</c> defaults to
            <c>undefined</c>. However, if the foreign key is set to
            something else, it causes the default values of the
            other fragmentation properties to be the same values as
            the actual fragmentation properties of the foreign table.</p>
        </item>
        <tag><c>{hash_module, Atom}</c></tag>
        <item>
          <p>Enables definition of an alternative hashing scheme.
            The module must implement the
            <seealso marker="mnesia_frag_hash">mnesia_frag_hash</seealso>
            callback behavior. This property can explicitly be set at
            table creation. Default is <c>mnesia_frag_hash</c>.</p>
          <p>Older tables, that were created before the concept of
            user-defined hash modules was introduced, use module
            <c>mnesia_frag_old_hash</c> to be backwards compatible.
            <c>mnesia_frag_old_hash</c> still uses the poor
            deprecated function <c>erlang:hash/1</c>.</p>
        </item>
        <tag><c>{hash_state, Term}</c></tag>
        <item>
          <p>Enables a table-specific parameterization of a
            generic hash module. This property can explicitly be set
            at table creation. Default is <c>undefined</c>.</p>
          <code type="none"><![CDATA[
Eshell V4.7.3.3  (abort with ^G)
(a@sam)1> mnesia:start().
ok
(a@sam)2> PrimProps = [{n_fragments, 7}, {node_pool, [node()]}].
[{n_fragments,7},{node_pool,[a@sam]}]
(a@sam)3> mnesia:create_table(prim_dict, 
                              [{frag_properties, PrimProps},
                               {attributes,[prim_key,prim_val]}]).
{atomic,ok}
(a@sam)4> SecProps = [{foreign_key, {prim_dict, sec_val}}].
[{foreign_key,{prim_dict,sec_val}}]
(a@sam)5> mnesia:create_table(sec_dict, 
                              [{frag_properties, SecProps},
(a@sam)5>                      {attributes, [sec_key, sec_val]}]).
{atomic,ok}
(a@sam)6> Write = fun(Rec) -> mnesia:write(Rec) end.
#Fun<erl_eval>
(a@sam)7> PrimKey = 11.
11
(a@sam)8> SecKey = 42.
42
(a@sam)9> mnesia:activity(sync_dirty, Write,
                          [{prim_dict, PrimKey, -11}], mnesia_frag).
ok
(a@sam)10> mnesia:activity(sync_dirty, Write,
                           [{sec_dict, SecKey, PrimKey}], mnesia_frag).
ok
(a@sam)11> mnesia:change_table_frag(prim_dict, {add_frag, [node()]}).
{atomic,ok}
(a@sam)12> SecRead = fun(PrimKey, SecKey) ->
               mnesia:read({sec_dict, PrimKey}, SecKey, read) end.
#Fun<erl_eval>
(a@sam)13> mnesia:activity(transaction, SecRead,
                           [PrimKey, SecKey], mnesia_frag).
[{sec_dict,42,11}]
(a@sam)14> Info = fun(Tab, Item) -> mnesia:table_info(Tab, Item) end.
#Fun<erl_eval>
(a@sam)15> mnesia:activity(sync_dirty, Info,
                           [prim_dict, frag_size], mnesia_frag).
[{prim_dict,0},
 {prim_dict_frag2,0},
 {prim_dict_frag3,0},
 {prim_dict_frag4,1},
 {prim_dict_frag5,0},
 {prim_dict_frag6,0},
 {prim_dict_frag7,0},
 {prim_dict_frag8,0}]
(a@sam)16> mnesia:activity(sync_dirty, Info,
                           [sec_dict, frag_size], mnesia_frag).
[{sec_dict,0},
 {sec_dict_frag2,0},
 {sec_dict_frag3,0},
 {sec_dict_frag4,1},
 {sec_dict_frag5,0},
 {sec_dict_frag6,0},
 {sec_dict_frag7,0},
 {sec_dict_frag8,0}]
(a@sam)17>
          ]]></code>
        </item>
      </taglist>
    </section>

    <section>
      <title>Management of Fragmented Tables</title>
      <p>The function <c>mnesia:change_table_frag(Tab, Change)</c>
        is intended to be used for reconfiguration of fragmented
        tables. Argument <c>Change</c> is to have one of the
        following values:</p>
      <taglist>
        <tag><c>{activate, FragProps}</c></tag>
        <item>
          <p>Activates the fragmentation properties of an
            existing table. <c>FragProps</c> is either to contain
            <c>{node_pool, Nodes}</c> or be empty.</p>
        </item>
        <tag><c>deactivate</c></tag>
        <item>
          <p>Deactivates the fragmentation properties of a
            table. The number of fragments must be <c>1</c>. No other
            table can refer to this table in its foreign key.</p>
        </item>
        <tag><c>{add_frag, NodesOrDist}</c></tag>
        <item>
          <p>Adds a fragment to a fragmented table. All
            records in one of the old fragments are rehashed and
            about half of them are moved to the new (last)
            fragment. All other fragmented tables, which refer to this
            table in their foreign key, automatically get a new
            fragment. Also, their records are dynamically
            rehashed in the same manner as for the main table.</p>
          <p>Argument <c>NodesOrDist</c> can either be a list of
            nodes or the result from the function
            <seealso marker="mnesia#table_info/2">mnesia:table_info(Tab, frag_dist)</seealso>.
            Argument <c>NodesOrDist</c> is
            assumed to be a sorted list with the best nodes to
            host new replicas first in the list. The new fragment
            gets the same number of replicas as the first
            fragment (see <c>n_ram_copies</c>, <c>n_disc_copies</c>,
            and <c>n_disc_only_copies</c>). The <c>NodesOrDist</c>
            list must at least contain one element for each
            replica that needs to be allocated.</p>
        </item>
        <tag><c>del_frag</c></tag>
        <item>
          <p>Deletes a fragment from a fragmented table. All
            records in the last fragment are  moved to one of the other
            fragments. All other fragmented tables, which refer to
            this table in their foreign key, automatically lose
            their last fragment. Also, their records are
            dynamically rehashed in the same manner as for the main
            table.</p>
        </item>
        <tag><c>{add_node, Node}</c></tag>
        <item>
          <p>Adds a node to <c>node_pool</c>. The new
            node pool affects the list returned from the function
            <seealso marker="mnesia#table_info/2">mnesia:table_info(Tab, frag_dist)</seealso>.
           </p>
        </item>
        <tag><c>{del_node, Node}</c></tag>
        <item>
          <p>Deletes a node from <c>node_pool</c>. The new
            node pool affects the list returned from the function
            <seealso marker="mnesia#table_info/2">mnesia:table_info(Tab, frag_dist)</seealso>.
          </p>
        </item>
      </taglist>
    </section>

    <section>
      <title>Extensions of Existing Functions</title>
      <p>The function
        <seealso marker="mnesia#create_table/2">mnesia:create_table/2</seealso>
        creates a brand new fragmented table, by setting table
        property <c>frag_properties</c> to some proper values.</p>
      <p>The function
        <seealso marker="mnesia#delete_table/1">mnesia:delete_table/1</seealso>
        deletes a fragmented table including all its
        fragments. There must however not exist any other fragmented
        tables that refer to this table in their foreign key.</p>
      <p>The function
        <seealso marker="mnesia#table_info/2">mnesia:table_info/2</seealso>
        now understands item <c>frag_properties</c>.</p>
      <p>If the function <c>mnesia:table_info/2</c> is started in
        the activity context of module <c>mnesia_frag</c>,
        information of several new items can be obtained:</p>
      <taglist>
        <tag><c>base_table</c></tag>
        <item>The name of the fragmented table</item>
        <tag><c>n_fragments</c></tag>
        <item>The actual number of fragments</item>
        <tag><c>node_pool</c></tag>
        <item>The pool of nodes</item>
        <tag><c>n_ram_copies</c></tag>
        <item></item>
        <tag><c>n_disc_copies</c></tag>
        <item></item>
        <tag><c>n_disc_only_copies</c></tag>
        <item>
          <p>The number of replicas with storage type <c>ram_copies</c>,
            <c>disc_copies</c>, and <c>disc_only_copies</c>,
            respectively. The actual values are dynamically derived
            from the first fragment. The first fragment serves as a
            protype. When the actual values need to be computed
            (for example, when adding new fragments) they are
            determined by counting the number of each replica for
            each storage type. This means that when the functions
            <seealso marker="mnesia#add_table_copy/3">mnesia:add_table_copy/3</seealso>,

            <seealso marker="mnesia#del_table_copy/2">mnesia:del_table_copy/2</seealso>,
            and
            <seealso marker="mnesia#change_table_copy_type/3">mnesia:change_table_copy_type/2</seealso> are applied on the
            first fragment, it affects the settings on
            <c>n_ram_copies</c>, <c>n_disc_copies</c>, and
            <c>n_disc_only_copies</c>.</p>
        </item>
        <tag><c>foreign_key</c></tag>
        <item>
          <p>The foreign key</p>
        </item>
        <tag><c>foreigners</c></tag>
        <item>
          <p>All other tables that refer to this table in
            their foreign key</p>
        </item>
        <tag><c>frag_names</c></tag>
        <item>
          <p>The names of all fragments</p>
        </item>
        <tag><c>frag_dist</c></tag>
        <item>
          <p>A sorted list of <c>{Node, Count}</c> tuples
            that are sorted in increasing <c>Count</c> order.
            <c>Count</c> is the total number of replicas that this
            fragmented table hosts on each <c>Node</c>. The list
            always contains at least all nodes in
            <c>node_pool</c>. Nodes that do not belong to
            <c>node_pool</c> are put last in the list even if
            their <c>Count</c> is lower.</p>
        </item>
        <tag><c>frag_size</c></tag>
        <item>
          <p>A list of <c>{Name, Size}</c> tuples, where
            <c>Name</c> is a fragment <c>Name</c>, and <c>Size</c> is
            how many records it contains</p>
        </item>
        <tag><c>frag_memory</c></tag>
        <item>
          <p>A list of <c>{Name, Memory}</c> tuples, where
            <c>Name</c> is a fragment <c>Name</c>, and <c>Memory</c> is
            how much memory it occupies</p>
        </item>
        <tag><c>size</c></tag>
        <item>
          <p>Total size of all fragments</p>
        </item>
        <tag><c>memory</c></tag>
        <item>
          <p>Total memory of all fragments</p>
        </item>
      </taglist>
    </section>

    <section>
      <title>Load Balancing</title>
      <p>There are several algorithms for distributing records
        in a fragmented table evenly over a
        pool of nodes. No one is best, it depends on the
        application needs. The following examples of
        situations need some attention:</p>
      <list type="bulleted">
       <item><c>permanent change of nodes</c>. When a new permanent
        <c>db_node</c> is introduced or dropped, it can be time to
        change the pool of nodes and redistribute the replicas
        evenly over the new pool of nodes. It can also be time to
        add or delete a fragment before the replicas are redistributed.
       </item>
       <item><c>size/memory threshold</c>. When the total size or
        total memory of a fragmented table (or a single
        fragment) exceeds some application-specific threshold, it
        can be time to add a new fragment dynamically to
        obtain a better distribution of records.
       </item>
       <item><c>temporary node down</c>. When a node temporarily goes
        down, it can be time to compensate some fragments with new
        replicas to keep the desired level of
        redundancy. When the node comes up again, it can be time to
        remove the superfluous replica.
       </item>
       <item><c>overload threshold</c>. When the load on some node
        exceeds some application-specific threshold, it can be time to
        either add or move some fragment replicas to nodes with lower
        load. Take extra care if the table has a foreign
        key relation to some other table. To avoid severe
        performance penalties, the same redistribution must be
        performed for all the related tables.
       </item>
      </list>
      <p>Use the function
        <c>mnesia:change_table_frag/2</c> to add new fragments
        and apply the usual schema manipulation functions (such as
        <seealso marker="mnesia#add_table_copy/3">mnesia:add_table_copy/3</seealso>,
        <seealso marker="mnesia#del_table_copy/2">mnesia:del_table_copy/2</seealso>,
        and
        <seealso marker="mnesia#change_table_copy_type/3">mnesia:change_table_copy_type/2</seealso>)
        on each fragment to perform the actual redistribution.</p>
    </section>
  </section>

  <section>
    <title>Local Content Tables</title>
    <p>Replicated tables have the same content on all nodes where
      they are replicated. However, it is sometimes advantageous to
      have tables, but different content on different nodes.</p>
    <p>If attribute <c>{local_content, true}</c> is specified when
      you create the table, the table resides on the nodes where you
      specify the table to exist, but the write operations on the
      table are only performed on the local copy.</p>
    <p>Furthermore, when the table is initialized at startup, the
      table is only initialized locally, and the table
      content is not copied from another node.</p>
  </section>

  <section>
    <title>Disc-Less Nodes</title>
    <p><c>Mnesia</c> can be run on nodes that do not have a disc.
      Replicas of <c>disc_copies</c> or <c>disc_only_copies</c> are
      not possible on such nodes. This is especially troublesome for
      the <c>schema</c> table, as <c>Mnesia</c> needs the schema
      to initialize itself.</p>
    <p>The schema table can, as other tables, reside on one or
      more nodes. The storage type of the schema table can either
      be <c>disc_copies</c> or <c>ram_copies</c>
      (but not <c>disc_only_copies</c>). At
      startup, <c>Mnesia</c> uses its schema to determine with which
      nodes it is to try to establish contact. If any
      other node is started already, the starting node
      merges its table definitions with the table definitions
      brought from the other nodes. This also applies to the
      definition of the schema table itself. Application
      parameter <c>extra_db_nodes</c> contains a list of nodes that
      <c>Mnesia</c> also is to establish contact with besides those
      found in the schema. Default is <c>[]</c> (empty list).</p>
    <p>Hence, when a disc-less node needs to find the schema
      definitions from a remote node on the network, this
      information must be supplied through application parameter
      <c>-mnesia extra_db_nodes NodeList</c>. Without this
      configuration parameter set, <c>Mnesia</c> starts as a single
      node system. Also, the function
      <seealso marker="mnesia#change_config/2">mnesia:change_config/2</seealso>
      can be used to assign a value to <c>extra_db_nodes</c> and force
      a connection after <c>Mnesia</c> has been started, that is,
      <c>mnesia:change_config(extra_db_nodes, NodeList)</c>.</p>
    <p>Application parameter <c>schema_location</c> controls where
      <c>Mnesia</c> searches for its schema. The parameter can be one
      of the following atoms:</p>
    <taglist>
      <tag><c>disc</c></tag>
      <item>
        <p>Mandatory disc. The schema is assumed to be located
          in the <c>Mnesia</c> directory. If the schema cannot be found,
          <c>Mnesia</c> refuses to start.</p>
      </item>
      <tag><c>ram</c></tag>
      <item>
        <p>Mandatory RAM. The schema resides in RAM
          only. At startup, a tiny new schema is generated. This
          default schema contains only the definition of the schema
          table and resides on the local node only. Since no other
          nodes are found in the default schema, configuration
          parameter <c>extra_db_nodes</c> must be used to let the
          node share its table definitions with other nodes. (Parameter
          <c>extra_db_nodes</c> can also be used on disc-full nodes.)</p>
      </item>
      <tag><c>opt_disc</c></tag>
      <item>
        <p>Optional disc. The schema can reside on either disc or
          RAM. If the schema is found on disc, <c>Mnesia</c> starts as
          a disc-full node (the storage type of the schema table is
          disc_copies). If no schema is found on disc, <c>Mnesia</c>
          starts as a disc-less node (the storage type of the schema
          table is <c>ram_copies</c>). The default for the
          application parameter is <c>opt_disc</c>.</p>
      </item>
    </taglist>
    <p>When <c>schema_location</c> is set to <c>opt_disc</c>, the
      function
      <seealso marker="mnesia#change_table_copy_type/3">mnesia:change_table_copy_type/3</seealso>
      can be used to change the storage type of the schema.
      This is illustrated as follows:</p>
    <pre>
        1> mnesia:start().
        ok
        2> mnesia:change_table_copy_type(schema, node(), disc_copies).
        {atomic, ok}</pre>
    <p>Assuming that the call to
      <seealso marker="mnesia#start/0">mnesia:start/0</seealso> does not
      find any schema to read on the disc, <c>Mnesia</c> starts
      as a disc-less node, and then change it to a node that
      use the disc to store the schema locally.</p>
  </section>

  <section>
    <title>More about Schema Management</title>
    <p>Nodes can be added to and removed from a <c>Mnesia</c> system.
      This can be done by adding a copy of the schema to those nodes.</p>
    <p>The functions
      <seealso marker="mnesia#add_table_copy/3">mnesia:add_table_copy/3</seealso>
      and
      <seealso marker="mnesia#del_table_copy/2">mnesia:del_table_copy/2</seealso>
      can be used to add and delete
      replicas of the schema table. Adding a node to the list of
      nodes where the schema is replicated affects the following:</p>
    <list type="bulleted">
      <item>It allows other tables to be replicated to this node.
      </item>
      <item>It causes <c>Mnesia</c> to try to contact the node at
        startup of disc-full nodes.
      </item>
    </list>
    <p>The function call <c>mnesia:del_table_copy(schema,
      mynode@host)</c> deletes node <c>mynode@host</c> from the
      <c>Mnesia</c> system. The call fails if <c>Mnesia</c> is running
      on <c>mynode@host</c>. The other <c>Mnesia</c> nodes never try to
      connect to that node again. Notice that if there is a disc resident
      schema on node <c>mynode@host</c>, the entire <c>Mnesia</c>
      directory is to be deleted. This is done with the function
      <seealso marker="mnesia#delete_schema/1">mnesia:delete_schema/1</seealso>.
      If <c>Mnesia</c> is started again
      on node <c>mynode@host</c> and the directory has not been
      cleared, the behavior of <c>Mnesia</c> is undefined.</p>
    <p>If the storage type of the schema is <c>ram_copies</c>,
      that is, a disc-less node, <c>Mnesia</c>
      does not use the disc on that particular node. The disc
      use is enabled by changing the storage type of table
      <c>schema</c> to <c>disc_copies</c>.</p>
    <p>New schemas are created explicitly with the function
      <seealso marker="mnesia#create_schema/1">mnesia:create_schema/1</seealso>
      or implicitly by starting
      <c>Mnesia</c> without a disc resident schema. Whenever
      a table (including the schema table) is created, it is
      assigned its own unique cookie. The schema table is not created
      with the function
      <seealso marker="mnesia#create_table/2">mnesia:create_table/2</seealso>
      as normal tables.</p>
    <p>At startup, <c>Mnesia</c> connects different nodes to each other,
      then they exchange table definitions with each other, and the table
      definitions are merged. During the merge procedure, <c>Mnesia</c>
      performs a sanity test to ensure that the table definitions are
      compatible with each other. If a table exists on several nodes,
      the cookie must be the same, otherwise <c>Mnesia</c> shut down one
      of the nodes. This unfortunate situation occurs if a table
      has been created on two nodes independently of each other while
      they were disconnected. To solve this, one of the tables
      must be deleted (as the cookies differ, it is regarded to be two
      different tables even if they have the same name).</p>
    <p>Merging different versions of the schema table does not
      always require the cookies to be the same. If the storage
      type of the schema table is <c>disc_copies</c>, the cookie is
      immutable, and all other <c>db_nodes</c> must have the same
      cookie. When the schema is stored as type <c>ram_copies</c>,
      its cookie can be replaced with a cookie from another node
      (<c>ram_copies</c> or <c>disc_copies</c>). The cookie replacement
      (during merge of the schema table definition) is performed each
      time a RAM node connects to another node.</p>
    <p>Further, the following applies:</p>
    <list type ="bulleted">
      <item><seealso marker="mnesia#system_info/1">mnesia:system_info(schema_location)</seealso>
        and
        <seealso marker="mnesia#system_info/1">mnesia:system_info(extra_db_nodes)</seealso>
        can be used to determine the actual values of <c>schema_location</c>
        and <c>extra_db_nodes</c>, respectively.
      </item>
      <item><seealso marker="mnesia#system_info/1">mnesia:system_info(use_dir)</seealso>
        can be used to determine whether <c>Mnesia</c> is actually
        using the <c>Mnesia</c> directory.
      </item>
      <item><c>use_dir</c> can be determined even before
        <c>Mnesia</c> is started.
     </item>
    </list>
    <p>The function <seealso marker="mnesia#info/0">mnesia:info/0</seealso>
      can now be used to print
      some system information even before <c>Mnesia</c> is started.
      When <c>Mnesia</c> is started, the function prints more
      information.</p>
    <p>Transactions that update the definition of a table
      requires that <c>Mnesia</c> is started on all nodes where the
      storage type of the schema is <c>disc_copies</c>. All replicas of
      the table on these nodes must also be loaded. There are a
      few exceptions to these availability rules:</p>
    <list type="bulleted">
      <item>Tables can be created and new replicas can be added
        without starting all the disc-full nodes.
      </item>
      <item>New replicas can be added before all other replicas of
        the table have been loaded, provided that at least one other
        replica is active.
      </item>
    </list>
  </section>

  <section>
    <marker id="event_handling"></marker>
    <title>Mnesia Event Handling</title>
    <p>System events and table events are the two event categories
      that <c>Mnesia</c> generates in various situations.</p>
    <p>A user process can subscribe on the events generated by
      <c>Mnesia</c>. The following two functions are provided:</p>
    <taglist>
      <tag><seealso marker="mnesia#subscribe/1">mnesia:subscribe(Event-Category)</seealso>
      </tag>
      <item>Ensures that a copy of all events of type
        <c>Event-Category</c> are sent to the calling process</item>
      <tag><seealso marker="mnesia#unsubscribe/1">mnesia:unsubscribe(Event-Category)</seealso>
      </tag>
      <item>Removes the subscription on events of type
        <c>Event-Category</c>
      </item>
    </taglist>
    <p><c>Event-Category</c> can be either of the following:</p>
    <list type="bulleted">
      <item>The atom <c>system</c>
      </item>
      <item>The atom <c>activity</c>
      </item>
      <item>The tuple <c>{table, Tab, simple}</c>
      </item>
      <item>The tuple <c>{table, Tab, detailed}</c>
      </item>
    </list>
    <p>The old event category <c>{table, Tab}</c> is the same
      event category as <c>{table, Tab, simple}</c>.</p>
    <p>The subscribe functions activate a subscription
      of events. The events are delivered as messages to the process
      evaluating the function
      <seealso marker="mnesia#subscribe/1">mnesia:subscribe/1</seealso>
      The syntax is as follows:</p>
    <list type="bulleted">
      <item><c>{mnesia_system_event, Event}</c> for system events
      </item>
      <item><c>{mnesia_activity_event, Event}</c> for activity events
      </item>
      <item><c>{mnesia_table_event, Event}</c> for table events
      </item>
    </list>
    <p>The event types are described in the next sections.</p>
    <p>All system events are subscribed by the <c>Mnesia</c>
      <c>gen_event</c> handler. The default <c>gen_event</c> handler
      is <c>mnesia_event</c>, but it can be changed by using
      application parameter <c>event_module</c>. The value of this
      parameter must be the name of a module implementing a complete
      handler, as specified by the
      <seealso marker="stdlib:gen_event">gen_event</seealso> module
      in <c>STDLIB</c>.</p>
    <p><seealso marker="mnesia#system_info/1">mnesia:system_info(subscribers)</seealso>
      and
      <seealso marker="mnesia#table_info/2">mnesia:table_info(Tab, subscribers)</seealso>
      can be used to determine which processes are subscribed to
      various events.</p>

    <section>
      <title>System Events</title>
      <p>The system events are as follows:</p>
      <taglist>
        <tag><c>{mnesia_up, Node}</c></tag>
        <item>Mnesia is started on a node. <c>Node</c> is the node
          name. By default this event is ignored.
        </item>
        <tag><c>{mnesia_down, Node}</c></tag>
        <item>Mnesia is stopped on a node. <c>Node</c> is the node
          name. By default this event is ignored.
        </item>
        <tag><c>{mnesia_checkpoint_activated, Checkpoint}</c></tag>
        <item>A checkpoint with the name <c>Checkpoint</c> is
          activated and the current node is involved in the
          checkpoint. Checkpoints can be activated explicitly with
          the function
          <seealso marker="mnesia#activate_checkpoint/1">mnesia:activate_checkpoint/1</seealso>
          or implicitly at
          backup, when adding table replicas, at internal transfer of
          data between nodes, and so on. By default this event is
          ignored.
        </item>
        <tag><c>{mnesia_checkpoint_deactivated, Checkpoint}</c></tag>
        <item>A checkpoint with the name <c>Checkpoint</c> is
          deactivated and the current node is involved in the
          checkpoint. Checkpoints can be deactivated explicitly with
          the function
          <seealso marker="mnesia#deactivate_checkpoint/1">mnesia:deactivate/1</seealso>
          or implicitly when the last
          replica of a table (involved in the checkpoint) becomes
          unavailable, for example, at node-down. By default this
          event is ignored.
        </item>
        <tag><c>{mnesia_overload, Details}</c></tag>
        <item><p><c>Mnesia</c> on the current node is
          overloaded and the subscriber is to take action.</p>
          <p>A typical overload situation occurs when the
            applications perform more updates on disc resident
            tables than <c>Mnesia</c> can handle. Ignoring
            this kind of overload can lead to a situation where
            the disc space is exhausted (regardless of the size of
            the tables stored on disc).</p>
          <p>Each update is appended to the transaction log and
            occasionally (depending on how it
            is configured) dumped to the tables files. The
            table file storage is more compact than the transaction
            log storage, especially if the same record is updated
            repeatedly. If the thresholds for dumping the
            transaction log are reached before the previous
            dump is finished, an overload event is triggered.</p>
          <p>Another typical overload situation is when the
            transaction manager cannot commit transactions at the
            same pace as the applications perform updates of
            disc resident tables. When this occurs, the message
            queue of the transaction manager continues to grow
            until the memory is exhausted or the load
            decreases.</p>
          <p>The same problem can occur for dirty updates. The overload
            is detected locally on the current node, but its cause can
            be on another node. Application processes can cause high
            load if any table resides on another node (replicated
            or not). By default this event
            is reported to <c>error_logger.</c></p>
        </item>
        <tag><c>{inconsistent_database, Context, Node}</c></tag>
        <item><c>Mnesia</c> regards the database as potential
          inconsistent and gives its applications a chance to
          recover from the inconsistency. For example, by installing a
          consistent backup as fallback and then restart the system.
          An alternative is to pick a <c>MasterNode</c> from
          <seealso marker="mnesia#system_info/1">mnesia:system_info(db_nodes)</seealso>
          and invoke
          <seealso marker="mnesia#set_master_nodes/1">mnesia:set_master_node([MasterNode])</seealso>.
          By default an error is reported to <c>error_logger</c>.
        </item>
        <tag><c>{mnesia_fatal, Format, Args, BinaryCore}</c></tag>
        <item>
          <p><c>Mnesia</c> detected a fatal error and
            terminates soon. The fault reason is explained in
            <c>Format</c> and <c>Args</c>, which can be given as input
            to <c>io:format/2</c> or sent to <c>error_logger</c>. By
            default it is sent to <c>error_logger</c>.</p>
          <p><c>BinaryCore</c> is a binary containing a summary of the
            <c>Mnesia</c> internal state at the time when the fatal
            error was detected. By default the binary is written to a
            unique filename on the current directory. On RAM nodes, the
            core is ignored.</p>
        </item>
        <tag><c>{mnesia_info, Format, Args}</c></tag>
        <item><c>Mnesia</c> detected something that can be of
          interest when debugging the system. This is explained in
          <c>Format</c> and <c>Args</c>, which can appear as input
          to <c>io:format/2</c> or sent to <c>error_logger</c>. By
          default this event is printed with <c>io:format/2</c>.
        </item>
        <tag><c>{mnesia_error, Format, Args}</c></tag>
        <item><c>Mnesia</c> has detected an error. The fault reason is
          explained in <c>Format</c> and <c>Args</c>, which can be
          given as input to <c>io:format/2</c> or sent to
          <c>error_logger</c>. By default this event is reported to
          <c>error_logger</c>.
        </item>
        <tag><c>{mnesia_user, Event}</c></tag>
        <item>An application started the function
          <seealso marker="mnesia#report_event/1">mnesia:report_event(Event)</seealso>.
          <c>Event</c> can be
          any Erlang data structure. When tracing a system of
          <c>Mnesia</c> applications, it is useful to be able to
          interleave own events of <c>Mnesia</c> with application-related
          events that give information about the application context.
          Whenever the application starts with a new and demanding
          <c>Mnesia</c> activity, or enters a new and interesting
          phase in its execution, it can be a good idea to use
          <c>mnesia:report_event/1</c>.
        </item>
      </taglist>
    </section>

    <section>
      <title>Activity Events</title>
      <p>Currently, there is only one type of activity event:</p>
      <taglist>
       <tag><c>{complete, ActivityID}</c></tag>
       <item>
         <p>This event occurs when a transaction that caused a modification
           to the database is completed. It is useful for determining when
           a set of table events (see the next section), caused by a given
           activity, have been sent. Once this event is received, it is
           guaranteed that no further table events with the same
           <c>ActivityID</c> will be received. Notice that this event can
           still be received even if no table events with a corresponding
           <c>ActivityID</c> were received, depending on
           the tables to which the receiving process is subscribed.</p>
         <p>Dirty operations always contain only one update and thus no
           activity event is sent.</p>
       </item>
     </taglist>
    </section>

    <section>
      <title>Table Events</title>
      <p>Table events are events related to table updates. There are
        two types of table events, simple and detailed.</p>
      <p>The <em>simple table events</em> are tuples like
        <c>{Oper, Record, ActivityId}</c>, where:</p>
      <list type="bulleted">
        <item><c>Oper</c> is the operation performed.
        </item>
        <item><c>Record</c> is the record involved in the operation.
        </item>
        <item><c>ActivityId</c> is the identity of the transaction
          performing the operation.
        </item>
      </list>
      <p>Notice that the record name is the table name even when
        <c>record_name</c> has another setting.</p>
      <p>The table-related events that can occur are as follows:</p>
      <taglist>
        <tag><c>{write, NewRecord, ActivityId}</c></tag>
        <item>A new record has been written. <c>NewRecord</c> contains
          the new record value.
        </item>
        <tag><c>{delete_object, OldRecord, ActivityId}</c></tag>
        <item>A record has possibly been deleted with
          <seealso marker="mnesia#delete_object/1">mnesia:delete_object/1</seealso>.
          <c>OldRecord</c>
          contains the value of the old record, as stated as argument
          by the application. Notice that other records with the same
          key can remain in the table if it is of type <c>bag</c>.
        </item>
        <tag><c>{delete, {Tab, Key}, ActivityId}</c></tag>
        <item>One or more records have possibly been deleted.
          All records with the key <c>Key</c> in the table
          <c>Tab</c> have been deleted.
        </item>
      </taglist>
      <p>The <em>detailed table events</em> are tuples like
        <c>{Oper, Table, Data, [OldRecs], ActivityId}</c>, where:</p>
      <list type="bulleted">
        <item><c>Oper</c> is the operation performed.
        </item>
        <item><c>Table</c> is the table involved in the operation.
        </item>
        <item><c>Data</c> is the record/OID written/deleted.
        </item>
        <item><c>OldRecs</c> is the contents before the operation.
        </item>
        <item><c>ActivityId</c> is the identity of the transaction
          performing the operation.
        </item>
      </list>
      <p>The table-related events that can occur are as follows:</p>
      <taglist>
        <tag><c>{write, Table, NewRecord, [OldRecords], ActivityId}</c></tag>
        <item>A new record has been written. <c>NewRecord</c> contains
          the new record value and <c>OldRecords</c> contains the
          records before the operation is performed. Notice that the
          new content depends on the table type.
        </item>
        <tag><c>{delete, Table, What, [OldRecords], ActivityId}</c></tag>
        <item>Records have possibly been deleted. <c>What</c> is
          either <c>{Table, Key}</c> or a record
          <c>{RecordName, Key, ...}</c> that was deleted. Notice
          that the new content depends on the table type.
        </item>
      </taglist>
    </section>
  </section>

  <section>
    <title>Debugging Mnesia Applications</title>
    <p>Debugging a <c>Mnesia</c> application can be difficult
      for various reasons, primarily related
      to difficulties in understanding how the transaction
      and table load mechanisms work. Another source of
      confusion can be the semantics of nested transactions.</p>
    <p>The debug level of <c>Mnesia</c> is set by calling the function
     <seealso marker="mnesia#set_debug_level/1">mnesia:set_debug_level(Level)</seealso>,
      where <c>Level</c>is one of the following:</p>
    <taglist>
      <tag><c>none</c></tag>
      <item>No trace outputs. This is the default.
      </item>
      <tag><c>verbose</c></tag>
      <item>Activates tracing of important debug events. These
        events generate <c>{mnesia_info, Format, Args}</c>
        system events. Processes can subscribe to these events with
        the function
        <seealso marker="mnesia#subscribe/1">mnesia:subscribe/1</seealso>.
        The events are always sent to the <c>Mnesia</c> event handler.
      </item>
      <tag><c>debug</c></tag>
      <item>Activates all events at the verbose level plus
          traces of all debug events. These debug events generate
          <c>{mnesia_info, Format, Args}</c> system events. Processes
          can subscribe to these events with <c>mnesia:subscribe/1</c>.
          The events are always sent to the <c>Mnesia</c> event handler.
          On this debug level, the <c> Mnesia</c> event handler starts
          subscribing to updates in the schema table.
      </item>
      <tag><c>trace</c></tag>
      <item>Activates all events at the debug level. On this
          level, the <c>Mnesia</c> event handler starts subscribing to
          updates on all <c>Mnesia</c> tables. This level is intended
          only for debugging small toy systems, as many large
          events can be generated.
      </item>
      <tag><c>false</c></tag>
      <item>An alias for none.
      </item>
      <tag><c>true</c></tag>
      <item>An alias for debug.
      </item>
    </taglist>
    <p>The debug level of <c>Mnesia</c> itself is also an application
      parameter, making it possible to start an Erlang system
      to turn on <c>Mnesia</c> debug in the initial
      startup phase by using the following code:</p>
    <pre>
      % erl -mnesia debug verbose</pre>
  </section>

  <section>
    <title>Concurrent Processes in Mnesia</title>
    <p>Programming concurrent Erlang systems is the subject of
      a separate book. However, it is worthwhile to draw attention to
      the following features, which permit concurrent processes to
      exist in a <c>Mnesia</c> system:</p>
    <list type="bulleted">
    <item><p>A group of functions or processes can be called within a
      transaction. A transaction can include statements that read,
      write, or delete data from the DBMS. Many such
      transactions can run concurrently, and the programmer does not
      need to explicitly synchronize the processes that manipulate
      the data.</p>
      <p>All programs accessing the database through the
      transaction system can be written as if they had sole access to
      the data. This is a desirable property, as all
      synchronization is taken care of by the transaction handler. If
      a program reads or writes data, the system ensures that no other
      program tries to manipulate the same data at the same time.</p>
    </item>
    <item>Tables can be moved or deleted, and the layout of a table
      can be reconfigured in various ways. An important aspect of
      the implementation of these functions is that user programs
      can continue to use a table while it
      is being reconfigured. For example, it is possible to move a
      table and perform write operations to the table at the same
      time. This is important for many applications that require
      continuously available services. For more information, see
      <seealso marker="Mnesia_chap4#trans_prop">Transactions and Other Access Contexts</seealso>.
    </item>
    </list>
  </section>

  <section>
    <title>Prototyping</title>
    <p>If and when you would like to start and manipulate
      <c>Mnesia</c>, it is often easier to write the definitions and
      data into an ordinary text file.
      Initially, no tables and no data exist, or which
      tables are required. At the initial stages of prototyping, it
      is prudent to write all data into one file, process that
      file, and have the data in the file inserted into the database.
      <c>Mnesia</c> can be initialized with data read from a text file.
      The following two functions can be used to work with text
      files.</p>
    <list type="bulleted">
      <item>
        <seealso marker="mnesia#load_textfile/1">mnesia:load_textfile(Filename)</seealso>
        loads a series of local table definitions and data found in the
        file into <c>Mnesia</c>. This function also starts <c>Mnesia</c>
        and possibly creates a new schema. The function operates
        on the local node only.
      </item>
      <item>
        <seealso marker="mnesia#dump_to_textfile/1">mnesia:dump_to_textfile(Filename)</seealso>
        dumps all local
        tables of a <c>Mnesia</c> system into a text file, which
        can be edited (with a normal text editor) and later reloaded.
      </item>
    </list>
    <p>These functions are much slower than the ordinary store and
      load functions of <c>Mnesia</c>. However, this is mainly intended
      for minor experiments and initial prototyping. The major
      advantage of these functions is that they are easy to use.</p>
    <p>The format of the text file is as follows:</p>
    <pre>
      {tables, [{Typename, [Options]},
      {Typename2 ......}]}.
      
      {Typename, Attribute1, Attribute2 ....}.
      {Typename, Attribute1, Attribute2 ....}.</pre>
    <p><c>Options</c> is a list of <c>{Key,Value}</c> tuples conforming
      to the options that you can give to
      <seealso marker="mnesia#create_table/2">mnesia:create_table/2</seealso>.
    </p>
    <p>For example, to start playing with a small database for healthy
      foods, enter the following data into file <c>FRUITS</c>:</p>
    <codeinclude file="FRUITS" tag="%0" type="erl"></codeinclude>
    <p>The following session with the Erlang shell shows how
      to load the <c>FRUITS</c> database:</p>
    <pre><![CDATA[
      % erl
      Erlang (BEAM) emulator version 4.9
 
      Eshell V4.9  (abort with ^G)
      1> mnesia:load_textfile("FRUITS").
      New table fruit
      New table vegetable
      {atomic,ok}
      2> mnesia:info().
      ---> Processes holding locks <--- 
      ---> Processes waiting for locks <--- 
      ---> Pending (remote) transactions <--- 
      ---> Active (local) transactions <---
      ---> Uncertain transactions <--- 
      ---> Active tables <--- 
      vegetable      : with 2 records occuping 299 words of mem 
      fruit          : with 2 records occuping 291 words of mem 
      schema         : with 3 records occuping 401 words of mem 
      ===> System info in version "1.1", debug level = none <===
      opt_disc. Directory "/var/tmp/Mnesia.nonode@nohost" is used.
      use fallback at restart = false
      running db nodes = [nonode@nohost]
      stopped db nodes = [] 
      remote           = []
      ram_copies       = [fruit,vegetable]
      disc_copies      = [schema]
      disc_only_copies = []
      [{nonode@nohost,disc_copies}] = [schema]
      [{nonode@nohost,ram_copies}] = [fruit,vegetable]
      3 transactions committed, 0 aborted, 0 restarted, 2 logged to disc
      0 held locks, 0 in queue; 0 local transactions, 0 remote
      0 transactions waits for other nodes: []
      ok
      3> 
    ]]></pre>
    <p>It can be seen that the DBMS was initiated from a
      regular text file.</p>
  </section>

  <section>
    <title>Object-Based Programming with Mnesia</title>
    <p>The <c>Company</c> database, introduced in
      <seealso marker="Mnesia_chap2#getting_started">Getting Started</seealso>,
      has three tables that store records (<c>employee</c>,
      <c>dept</c>, <c>project</c>), and three tables that store
      relationships (<c>manager</c>, <c>at_dep</c>, <c>in_proj</c>).
      This is a normalized data model, which has some advantages over
      a non-normalized data model.</p>
    <p>It is more efficient to do a
      generalized search in a normalized database. Some operations are
      also easier to perform on a normalized data model. For example,
      one project can easily be removed, as the following example
      illustrates:</p>
    <codeinclude file="company.erl" tag="%13" type="erl"></codeinclude>
    <p>In reality, data models are seldom fully normalized. A
      realistic alternative to a normalized database model would be
      a data model that is not even in first normal form. <c>Mnesia</c>
      is suitable for applications such as telecommunications,
      because it is easy to organize data in a flexible manner. A
      <c>Mnesia</c> database is always organized as a set of tables.
      Each table is filled with rows, objects, and records.
      What sets <c>Mnesia</c> apart is that individual fields in
      a record can contain any type of
      compound data structures. An individual field in a record can
      contain lists, tuples, functions, and even record code.</p>
    <p>Many telecommunications applications have unique requirements
      on lookup times for certain types of records. If the <c>Company</c>
      database had been a part of a telecommunications system, it
      could be to minimize the lookup time of an employee
      <em>together</em> with a list of the projects the employee is
      working on. If this is the case, a drastically different data model
      without direct relationships can be chosen. You would then have
      only the records themselves, and different records could contain
      either direct references to other records, or contain other
      records that are not part of the <c>Mnesia</c> schema.</p>
    <p>The following record definitions can be created:</p>
    <codeinclude file="company_o.hrl" tag="%0" type="erl"></codeinclude>
    <p>A record that describes an employee can look as follows:</p>
    <pre>
        Me = #employee{emp_no= 104732,
        name = klacke,
        salary = 7,
        sex = male,
        phone = 99586,
        room_no = {221, 015},
        dept = 'B/SFR',
        projects = [erlang, mnesia, otp],
        manager = 114872},</pre>
    <p>This model has only three different tables, and the employee
      records contain references to other records. The record has the
      following references:</p>
    <list type="bulleted">
      <item><c>'B/SFR'</c> refers to a <c>dept</c> record.
      </item>
      <item><c>[erlang, mnesia, otp]</c> is a list of three
       direct references to three different <c>projects</c> records.
      </item>
      <item><c>114872</c> refers to another employee record.
      </item>
    </list>
    <p>The <c>Mnesia</c> record identifiers (<c>{Tab, Key}</c>) can
      also be used as references. In this case, attribute <c>dept</c>
      would be set to value <c>{dept, 'B/SFR'}</c> instead of
      <c>'B/SFR'</c>.</p>
    <p>With this data model, some operations execute considerably
      faster than they do with the normalized data model in the
      <c>Company</c> database. However, some other operations
      become much more complicated. In particular, it becomes more
      difficult to ensure that records do not contain dangling
      pointers to other non-existent, or deleted, records.</p>
    <p>The following code exemplifies a search with a non-normalized
      data model. To find all employees at department <c>Dep</c> with
      a salary higher than <c>Salary</c>, use the following code:</p>
    <codeinclude file="company_o.erl" tag="%9" type="erl"></codeinclude>
    <p>This code is easier to write and to understand, and it
      also executes much faster.</p>
    <p>It is easy to show examples of code that executes faster if
      a non-normalized data model is used, instead of a normalized
      model. The main reason is that fewer tables are required.
      Therefore, data from different tables can more easily be
      combined in join operations. In the previous example, the
      function <c>get_emps/2</c> is transformed from a join operation
      into a simple query, which consists of a selection and a
      projection on one single table.</p>
  </section>
</chapter>