postgresql/doc/src/sgml/wal.sgml

<!-- $PostgreSQL: pgsql/doc/src/sgml/wal.sgml,v 1.27 2004/01/06 17:26:23 neilc Exp $ -->

<chapter id="wal">
 <title>Write-Ahead Logging (<acronym>WAL</acronym>)</title>

 <indexterm zone="wal">
  <primary>WAL</primary>
 </indexterm>

 <indexterm>
  <primary>transaction log</primary>
  <see>WAL</see>
 </indexterm>

  <para>
   <firstterm>Write-Ahead Logging</firstterm> (<acronym>WAL</acronym>)
   is a standard approach to transaction logging.  Its detailed
   description may be found in most (if not all) books about
   transaction processing. Briefly, <acronym>WAL</acronym>'s central
   concept is that changes to data files (where tables and indexes
   reside) must be written only after those changes have been logged,
   that is, when log records describing the changes have been flushed
   to permanent storage. If we follow this procedure, we do not need
   to flush data pages to disk on every transaction commit, because we
   know that in the event of a crash we will be able to recover the
   database using the log: any changes that have not been applied to
   the data pages will first be redone from the log records (this is
   roll-forward recovery, also known as REDO) and then changes made by
   uncommitted transactions will be removed from the data pages
   (roll-backward recovery, UNDO).
  </para>

  <sect1 id="wal-benefits-now">
   <title>Benefits of <acronym>WAL</acronym></title>

   <indexterm zone="wal-benefits-now">
    <primary>fsync</primary>
   </indexterm>

   <para>
    The first obvious benefit of using <acronym>WAL</acronym> is a
    significantly reduced number of disk writes, since only the log
    file needs to be flushed to disk at the time of transaction
    commit; in multiuser environments, commits of many transactions
    may be accomplished with a single <function>fsync()</function> of
    the log file. Furthermore, the log file is written sequentially,
    and so the cost of syncing the log is much less than the cost of
    flushing the data pages.
   </para>

   <para>
    The next benefit is consistency of the data pages. The truth is
    that, before <acronym>WAL</acronym>,
    <productname>PostgreSQL</productname> was never able to guarantee
    consistency in the case of a crash.  Before
    <acronym>WAL</acronym>, any crash during writing could result in:

    <orderedlist>
     <listitem>
      <simpara>index rows pointing to nonexistent table rows</simpara>
     </listitem>

     <listitem>
      <simpara>index rows lost in split operations</simpara>
     </listitem>

     <listitem>
      <simpara>totally corrupted table or index page content, because
      of partially written data pages</simpara>
     </listitem>
    </orderedlist>

    Problems with indexes (problems 1 and 2) could possibly have been
    fixed by additional <function>fsync()</function> calls, but it is
    not obvious how to handle the last case without
    <acronym>WAL</acronym>; <acronym>WAL</acronym> saves the entire data
    page content in the log if that is required to ensure page
    consistency for after-crash recovery.
   </para>
  </sect1>
  
  <sect1 id="wal-benefits-later">
   <title>Future Benefits</title>

   <para>
    The UNDO operation is not implemented. This means that changes
    made by aborted transactions will still occupy disk space and that
    a permanent <filename>pg_clog</filename> file to hold
    the status of transactions is still needed, since
    transaction identifiers cannot be reused. Once UNDO is implemented,
    <filename>pg_clog</filename> will no longer be required to be
    permanent; it will be possible to remove
    <filename>pg_clog</filename> at shutdown. (However, the urgency of
    this concern has decreased greatly with the adoption of a segmented
    storage method for <filename>pg_clog</filename>: it is no longer
    necessary to keep old <filename>pg_clog</filename> entries around
    forever.)
   </para>

   <para>
    With UNDO, it will also be possible to implement
    <firstterm>savepoints</firstterm><indexterm><primary>savepoint</></> to allow partial rollback of
    invalid transaction operations (parser errors caused by mistyping
    commands, insertion of duplicate primary/unique keys and so on)
    with the ability to continue or commit valid operations made by
    the transaction before the error.  At present, any error will
    invalidate the whole transaction and require a transaction abort.
   </para>

   <para>
    <acronym>WAL</acronym> offers the opportunity for a new method for
    database on-line backup and restore (<acronym>BAR</acronym>).  To
    use this method, one would have to make periodic saves of data
    files to another disk, a tape or another host and also archive the
    <acronym>WAL</acronym> log files.  The database file copy and the
    archived log files could be used to restore just as if one were
    restoring after a crash. Each time a new database file copy was
    made the old log files could be removed.  Implementing this
    facility will require the logging of data file and index creation
    and deletion; it will also require development of a method for
    copying the data files (operating system copy commands are not
    suitable).
   </para>

   <para>
    A difficulty standing in the way of realizing these benefits is that
    they require saving <acronym>WAL</acronym> entries for considerable
    periods of time (e.g., as long as the longest possible transaction if
    transaction UNDO is wanted). The present <acronym>WAL</acronym>
    format is extremely bulky since it includes many disk page
    snapshots. This is not a serious concern at present, since the
    entries only need to be kept for one or two checkpoint intervals;
    but to achieve these future benefits some sort of compressed
    <acronym>WAL</acronym> format will be needed.
   </para>
  </sect1>

 <sect1 id="wal-configuration">
  <title><acronym>WAL</acronym> Configuration</title>

  <para>
   There are several <acronym>WAL</acronym>-related configuration parameters that
   affect database performance. This section explains their use.
   Consult <xref linkend="runtime-config"> for details about setting
   configuration parameters.
  </para>

  <para>
   <firstterm>Checkpoints</firstterm><indexterm><primary>checkpoint</></> are points in the sequence of
   transactions at which it is guaranteed that the data files have
   been updated with all information logged before the checkpoint.  At
   checkpoint time, all dirty data pages are flushed to disk and a
   special checkpoint record is written to the log file. As result, in
   the event of a crash, the recoverer knows from what record in the
   log (known as the redo record) it should start the REDO operation,
   since any changes made to data files before that record are already
   on disk. After a checkpoint has been made, any log segments written
   before the redo records are no longer needed and can be recycled or
   removed. (When <acronym>WAL</acronym>-based <acronym>BAR</acronym> is
   implemented, the log segments would be archived before being recycled
   or removed.)
  </para>

  <para>
   The server spawns a special process every so often
   to create the next checkpoint.  A checkpoint is created every
   <varname>checkpoint_segments</varname> log segments, or every
   <varname>checkpoint_timeout</varname> seconds, whichever comes first.
   The default settings are 3 segments and 300 seconds respectively.
   It is also possible to force a checkpoint by using the SQL command
   <command>CHECKPOINT</command>.
  </para>

  <para>
   Reducing <varname>checkpoint_segments</varname> and/or
   <varname>checkpoint_timeout</varname> causes checkpoints to be done
   more often. This allows faster after-crash recovery (since less work
   will need to be redone). However, one must balance this against the
   increased cost of flushing dirty data pages more often. In addition,
   to ensure data page consistency, the first modification of a data
   page after each checkpoint results in logging the entire page
   content. Thus a smaller checkpoint interval increases the volume of
   output to the log, partially negating the goal of using a smaller
   interval, and in any case causing more disk I/O.
  </para>

  <para>
   There will be at least one 16 MB segment file, and will normally
   not be more than 2 * <varname>checkpoint_segments</varname> + 1
   files.  You can use this to estimate space requirements for <acronym>WAL</acronym>.
   Ordinarily, when old log segment files are no longer needed, they
   are recycled (renamed to become the next segments in the numbered
   sequence). If, due to a short-term peak of log output rate, there
   are more than 2 * <varname>checkpoint_segments</varname> + 1
   segment files, the unneeded segment files will be deleted instead
   of recycled until the system gets back under this limit.
  </para>

  <para>
   There are two commonly used <acronym>WAL</acronym> functions:
   <function>LogInsert</function> and <function>LogFlush</function>.
   <function>LogInsert</function> is used to place a new record into
   the <acronym>WAL</acronym> buffers in shared memory. If there is no
   space for the new record, <function>LogInsert</function> will have
   to write (move to kernel cache) a few filled <acronym>WAL</acronym>
   buffers. This is undesirable because <function>LogInsert</function>
   is used on every database low level modification (for example,
   row insertion) at a time when an exclusive lock is held on
   affected data pages, so the operation needs to be as fast as
   possible.  What is worse, writing <acronym>WAL</acronym> buffers may
   also force the creation of a new log segment, which takes even more
   time. Normally, <acronym>WAL</acronym> buffers should be written
   and flushed by a <function>LogFlush</function> request, which is
   made, for the most part, at transaction commit time to ensure that
   transaction records are flushed to permanent storage. On systems
   with high log output, <function>LogFlush</function> requests may
   not occur often enough to prevent <acronym>WAL</acronym> buffers
   being written by <function>LogInsert</function>. On such systems
   one should increase the number of <acronym>WAL</acronym> buffers by
   modifying the configuration parameter <varname>wal_buffers</varname>.
   The default number of <acronym>
   WAL</acronym> buffers is 8.  Increasing this value will 
   correspondingly increase shared memory usage.
  </para>

  <para>
   Checkpoints are fairly expensive because they force all dirty kernel
   buffers to disk using the operating system <literal>sync()</> call.
   Busy servers may fill checkpoint segment files too quickly,
   causing excessive checkpointing. If such forced checkpoints happen
   more frequently than <varname>checkpoint_warning</varname> seconds, 
   a message, will be output to the server logs recommending increasing 
   <varname>checkpoint_segments</varname>.
  </para>

  <para>
   The <varname>commit_delay</varname> parameter defines for how many
   microseconds the server process will sleep after writing a commit
   record to the log with <function>LogInsert</function> but before
   performing a <function>LogFlush</function>. This delay allows other
   server processes to add their commit records to the log so as to have all
   of them flushed with a single log sync. No sleep will occur if
   <varname>fsync</varname>
   is not enabled, nor if fewer than <varname>commit_siblings</varname>
   other sessions are currently in active transactions; this avoids
   sleeping when it's unlikely that any other session will commit soon.
   Note that on most platforms, the resolution of a sleep request is
   ten milliseconds, so that any nonzero <varname>commit_delay</varname>
   setting between 1 and 10000 microseconds would have the same effect.
   Good values for these parameters are not yet clear; experimentation
   is encouraged.
  </para>

  <para>
   The <varname>wal_sync_method</varname> parameter determines how
   <productname>PostgreSQL</productname> will ask the kernel to force
    <acronym>WAL</acronym> updates out to disk. 
   All the options should be the same as far as reliability goes,
   but it's quite platform-specific which one will be the fastest.
   Note that this parameter is irrelevant if <varname>fsync</varname>
   has been turned off.
  </para>

  <para>
   Enabling the <varname>wal_debug</varname> configuration parameter
   will result in each <function>LogInsert</function> and
   <function>LogFlush</function> <acronym>WAL</acronym> call being
   logged to the server log. This option may be replaced by a more
   general mechanism in the future.
  </para>
 </sect1>

 <sect1 id="wal-internals">
  <title>Internals</title>

  <para>
   <acronym>WAL</acronym> is automatically enabled; no action is
   required from the administrator except ensuring that the additional
   disk-space requirements of the <acronym>WAL</acronym> logs are met,
   and that any necessary tuning is done (see <xref
   linkend="wal-configuration">).
  </para>

  <para>
   <acronym>WAL</acronym> logs are stored in the directory
   <filename>pg_xlog</filename> under the data directory, as a set of
   segment files, each 16 MB in size.  Each segment is divided into 8
   kB pages. The log record headers are described in
   <filename>access/xlog.h</filename>; the record content is dependent
   on the type of event that is being logged.  Segment files are given
   ever-increasing numbers as names, starting at
   <filename>0000000000000000</filename>.  The numbers do not wrap, at
   present, but it should take a very long time to exhaust the
   available stock of numbers.
  </para>

  <para>
   The <acronym>WAL</acronym> buffers and control structure are in
   shared memory and are handled by the server child processes; they
   are protected by lightweight locks.  The demand on shared memory is
   dependent on the number of buffers.  The default size of the
   <acronym>WAL</acronym> buffers is 8 buffers of 8 kB each, or 64 kB
   total.
  </para>

  <para>
   It is of advantage if the log is located on another disk than the
   main database files.  This may be achieved by moving the directory
   <filename>pg_xlog</filename> to another location (while the server
   is shut down, of course) and creating a symbolic link from the
   original location in the main data directory to the new location.
  </para>

  <para>
   The aim of <acronym>WAL</acronym>, to ensure that the log is
   written before database records are altered, may be subverted by
   disk drives<indexterm><primary>disk drive</></> that falsely report a successful write to the kernel,
   when, in fact, they have only cached the data and not yet stored it
   on the disk.  A power failure in such a situation may still lead to
   irrecoverable data corruption.  Administrators should try to ensure
   that disks holding <productname>PostgreSQL</productname>'s
   <acronym>WAL</acronym> log files do not make such false reports.
  </para>

  <para>
   After a checkpoint has been made and the log flushed, the
   checkpoint's position is saved in the file
   <filename>pg_control</filename>. Therefore, when recovery is to be
   done, the server first reads <filename>pg_control</filename> and
   then the checkpoint record; then it performs the REDO operation by
   scanning forward from the log position indicated in the checkpoint
   record.  Because the entire content of data pages is saved in the
   log on the first page modification after a checkpoint, all pages
   changed since the checkpoint will be restored to a consistent
   state.
  </para>

  <para>
   Using <filename>pg_control</filename> to get the checkpoint
   position speeds up the recovery process, but to handle possible
   corruption of <filename>pg_control</filename>, we should actually
   implement the reading of existing log segments in reverse order --
   newest to oldest -- in order to find the last checkpoint.  This has
   not been implemented, yet.
  </para>
 </sect1>
</chapter>

<!-- Keep this comment at the end of the file
Local variables:
mode:sgml
sgml-omittag:nil
sgml-shorttag:t
sgml-minimize-attributes:nil
sgml-always-quote-attributes:t
sgml-indent-step:1
sgml-indent-data:t
sgml-parent-document:nil
sgml-default-dtd-file:"./reference.ced"
sgml-exposed-tags:nil
sgml-local-catalogs:("/usr/lib/sgml/catalog")
sgml-local-ecat-files:nil
End:
-->