codekingpro/portable-devtools

1<?xml version="1.0" encoding="UTF-8" standalone="no"?>2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>13.4. Data Consistency Checks at the Application Level</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><link rel="prev" href="explicit-locking.html" title="13.3. Explicit Locking" /><link rel="next" href="mvcc-serialization-failure-handling.html" title="13.5. Serialization Failure Handling" /></head><body id="docContent" class="container-fluid col-10"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">13.4. Data Consistency Checks at the Application Level</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="explicit-locking.html" title="13.3. Explicit Locking">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="mvcc.html" title="Chapter 13. Concurrency Control">Up</a></td><th width="60%" align="center">Chapter 13. Concurrency Control</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 16.3 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="mvcc-serialization-failure-handling.html" title="13.5. Serialization Failure Handling">Next</a></td></tr></table><hr /></div><div class="sect1" id="APPLEVEL-CONSISTENCY"><div class="titlepage"><div><div><h2 class="title" style="clear: both">13.4. Data Consistency Checks at the Application Level <a href="#APPLEVEL-CONSISTENCY" class="id_link">#</a></h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="applevel-consistency.html#SERIALIZABLE-CONSISTENCY">13.4.1. Enforcing Consistency with Serializable Transactions</a></span></dt><dt><span class="sect2"><a href="applevel-consistency.html#NON-SERIALIZABLE-CONSISTENCY">13.4.2. Enforcing Consistency with Explicit Blocking Locks</a></span></dt></dl></div><p>3    It is very difficult to enforce business rules regarding data integrity4    using Read Committed transactions because the view of the data is5    shifting with each statement, and even a single statement may not6    restrict itself to the statement's snapshot if a write conflict occurs.7   </p><p>8    While a Repeatable Read transaction has a stable view of the data9    throughout its execution, there is a subtle issue with using10    <acronym class="acronym">MVCC</acronym> snapshots for data consistency checks, involving11    something known as <em class="firstterm">read/write conflicts</em>.12    If one transaction writes data and a concurrent transaction attempts13    to read the same data (whether before or after the write), it cannot14    see the work of the other transaction.  The reader then appears to have15    executed first regardless of which started first or which committed16    first.  If that is as far as it goes, there is no problem, but17    if the reader also writes data which is read by a concurrent transaction18    there is now a transaction which appears to have run before either of19    the previously mentioned transactions.  If the transaction which appears20    to have executed last actually commits first, it is very easy for a21    cycle to appear in a graph of the order of execution of the transactions.22    When such a cycle appears, integrity checks will not work correctly23    without some help.24   </p><p>25    As mentioned in <a class="xref" href="transaction-iso.html#XACT-SERIALIZABLE" title="13.2.3. Serializable Isolation Level">Section 13.2.3</a>, Serializable26    transactions are just Repeatable Read transactions which add27    nonblocking monitoring for dangerous patterns of read/write conflicts.28    When a pattern is detected which could cause a cycle in the apparent29    order of execution, one of the transactions involved is rolled back to30    break the cycle.31   </p><div class="sect2" id="SERIALIZABLE-CONSISTENCY"><div class="titlepage"><div><div><h3 class="title">13.4.1. Enforcing Consistency with Serializable Transactions <a href="#SERIALIZABLE-CONSISTENCY" class="id_link">#</a></h3></div></div></div><p>32     If the Serializable transaction isolation level is used for all writes33     and for all reads which need a consistent view of the data, no other34     effort is required to ensure consistency.  Software from other35     environments which is written to use serializable transactions to36     ensure consistency should <span class="quote">“<span class="quote">just work</span>”</span> in this regard in37     <span class="productname">PostgreSQL</span>.38    </p><p>39     When using this technique, it will avoid creating an unnecessary burden40     for application programmers if the application software goes through a41     framework which automatically retries transactions which are rolled42     back with a serialization failure.  It may be a good idea to set43     <code class="literal">default_transaction_isolation</code> to <code class="literal">serializable</code>.44     It would also be wise to take some action to ensure that no other45     transaction isolation level is used, either inadvertently or to46     subvert integrity checks, through checks of the transaction isolation47     level in triggers.48    </p><p>49     See <a class="xref" href="transaction-iso.html#XACT-SERIALIZABLE" title="13.2.3. Serializable Isolation Level">Section 13.2.3</a> for performance suggestions.50    </p><div class="warning"><h3 class="title">Warning: Serializable Transactions and Data Replication</h3><p>51      This level of integrity protection using Serializable transactions52      does not yet extend to hot standby mode (<a class="xref" href="hot-standby.html" title="27.4. Hot Standby">Section 27.4</a>)53      or logical replicas.54      Because of that, those using hot standby or logical replication55      may want to use Repeatable Read and explicit locking on the primary.56     </p></div></div><div class="sect2" id="NON-SERIALIZABLE-CONSISTENCY"><div class="titlepage"><div><div><h3 class="title">13.4.2. Enforcing Consistency with Explicit Blocking Locks <a href="#NON-SERIALIZABLE-CONSISTENCY" class="id_link">#</a></h3></div></div></div><p>57     When non-serializable writes are possible,58     to ensure the current validity of a row and protect it against59     concurrent updates one must use <code class="command">SELECT FOR UPDATE</code>,60     <code class="command">SELECT FOR SHARE</code>, or an appropriate <code class="command">LOCK61     TABLE</code> statement.  (<code class="command">SELECT FOR UPDATE</code>62     and <code class="command">SELECT FOR SHARE</code> lock just the63     returned rows against concurrent updates, while <code class="command">LOCK64     TABLE</code> locks the whole table.)  This should be taken into65     account when porting applications to66     <span class="productname">PostgreSQL</span> from other environments.67    </p><p>68     Also of note to those converting from other environments is the fact69     that <code class="command">SELECT FOR UPDATE</code> does not ensure that a70     concurrent transaction will not update or delete a selected row.71     To do that in <span class="productname">PostgreSQL</span> you must actually72     update the row, even if no values need to be changed.73     <code class="command">SELECT FOR UPDATE</code> <span class="emphasis"><em>temporarily blocks</em></span>74     other transactions from acquiring the same lock or executing an75     <code class="command">UPDATE</code> or <code class="command">DELETE</code> which would76     affect the locked row, but once the transaction holding this lock77     commits or rolls back, a blocked transaction will proceed with the78     conflicting operation unless an actual <code class="command">UPDATE</code> of79     the row was performed while the lock was held.80    </p><p>81     Global validity checks require extra thought under82     non-serializable <acronym class="acronym">MVCC</acronym>.83     For example, a banking application might wish to check that the sum of84     all credits in one table equals the sum of debits in another table,85     when both tables are being actively updated.  Comparing the results of two86     successive <code class="literal">SELECT sum(...)</code> commands will not work reliably in87     Read Committed mode, since the second query will likely include the results88     of transactions not counted by the first.  Doing the two sums in a89     single repeatable read transaction will give an accurate picture of only the90     effects of transactions that committed before the repeatable read transaction91     started — but one might legitimately wonder whether the answer is still92     relevant by the time it is delivered.  If the repeatable read transaction93     itself applied some changes before trying to make the consistency check,94     the usefulness of the check becomes even more debatable, since now it95     includes some but not all post-transaction-start changes.  In such cases96     a careful person might wish to lock all tables needed for the check,97     in order to get an indisputable picture of current reality.  A98     <code class="literal">SHARE</code> mode (or higher) lock guarantees that there are no99     uncommitted changes in the locked table, other than those of the current100     transaction.101    </p><p>102     Note also that if one is relying on explicit locking to prevent concurrent103     changes, one should either use Read Committed mode, or in Repeatable Read104     mode be careful to obtain105     locks before performing queries.  A lock obtained by a106     repeatable read transaction guarantees that no other transactions modifying107     the table are still running, but if the snapshot seen by the108     transaction predates obtaining the lock, it might predate some now-committed109     changes in the table.  A repeatable read transaction's snapshot is actually110     frozen at the start of its first query or data-modification command111     (<code class="literal">SELECT</code>, <code class="literal">INSERT</code>,112     <code class="literal">UPDATE</code>, <code class="literal">DELETE</code>, or113     <code class="literal">MERGE</code>), so it is possible to obtain locks explicitly114     before the snapshot is frozen.115    </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="explicit-locking.html" title="13.3. Explicit Locking">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="mvcc.html" title="Chapter 13. Concurrency Control">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="mvcc-serialization-failure-handling.html" title="13.5. Serialization Failure Handling">Next</a></td></tr><tr><td width="40%" align="left" valign="top">13.3. Explicit Locking </td><td width="20%" align="center"><a accesskey="h" href="index.html" title="PostgreSQL 16.3 Documentation">Home</a></td><td width="40%" align="right" valign="top"> 13.5. Serialization Failure Handling</td></tr></table></div></body></html>