Pages

Wednesday, May 25, 2016

Troubleshooting ORA-1555

Troubleshooting ORA-1555


This document is intended to assist in finding tips and techniques to assist with finding solutions to ORA-1555 errors. The document will cover the following topics:

Concepts/Definitions
Diagnosing
Common Causes/Solutions
Concepts/Definitions

The ORA-1555 errors can happen when a query is unable to access enough undo to build
a copy of the data at the time the query started. Committed “versions” of blocks are
maintained along with newer uncommitted “versions” of those blocks so that queries can
access data as it existed in the database at the time of the query. These are referred to as
“consistent read” blocks and are maintained using Oracle undo management.

See Note 40689.1 - ORA-1555 "Snapshot too old" - Detailed Explanation for more about
these errors.
Diagnosing

Due to space limitations, it is not always feasible to keep undo blocks on hand for the life of the instance. Oracle Automatic Undo Management (AUM) helps to manage the time frame that undo blocks are stored. The time frame is the “retention” time for those blocks.

There are several ways to investigate the ORA-1555 error. In most cases, the error is a legitimate problem with getting to an undo block that has been overwritten due to the undo “retention” period having passed.

AUM will automatically tune up and down the “retention” period, but often space limitations or configuration of the undo tablespace will throttle back continuous increases to the “retention” period.

The error message is reported in the user session and often is not captured in the alert log. The user could see a message like

Using rollback segment functionality:
ORA-1555: snapshot too old (rollback segment too small)

or

Using AUM:
ORA-01555: snapshot too old: rollback segment number 9 with name "_SYSSMU9$" too small

If the error is captured in the alert.log, you would see something like

Tue May 26 16:16:57 2009
ORA-01555 caused by SQL statement below (SQL ID: 54yn3n36w24ft, Query Duration=922 sec, SCN: 0x0007.8a55f4e3)
Initial Investigation

Rollback Segments:

With Oracle 10g and later versions of Oracle, you can still use a Rollback Segments configuration. ORA-1555 errors in that environment still follow older guidelines as described in

Note 10579.1 - How many Rollback Segments to Have
Note 107085.1 - Tuning Rollback Segments
Note 69464.1 - Rollback Segment Configuration & Tips
Automatic Undo Management:

The database will be self tuning for undo when using Automatic Undo Management. This does not eliminate ORA-1555 completely, but does minimize ORA-1555 as long as there is adequate space in the undo tablespace and workloads tend to follow repeatable patterns. In some cases with periodic changes to workload (large data updates particularly with LOB data) the self tuning of undo can become aggressive and lead to undo issues.

Note 461480.1 - FAQ Automatic Undo Management (AUM) / System Managed Undo (SMU)
Note 135053.1 -How to Create a Database with Automatic Undo Management
Note 268870.1 - How to Shrink the datafile of Undo Tablespace
Note 231776.1 - How to switch a Database from Automatic Undo Management (AUM) back to using Rollback Segments
Note 296863.1 - How to Keep All UNDO Segments from Being Offlined in Oracle 10g - Fast Ramp-Up
LOB Issues:

Out-of-row LOB undo is maintained in the LOB segment. So the UNDO tablespace and undo retention is not associated with most LOB ORA-1555 issues. Instead the LOB column is created using either PCT_VERSION or RETENTION to manage how much space within blocks or time transpires before the LOB undo is overwritten. In environments with high updates, deletes on rows including LOBs, the chances of ORA-1555 on LOB undo is very high.

PCT_VERSION and RETENTION are not auto-tuned. To “tune” those configuration settings, you must change the values for PCT_VERSION or RETENTION. Changes to UNDO_RETENTION does not change LOB retention time frames.

Note 162345.1 - LOBS - Storage, Read-consistency and Rollback
Note 386341.1 - How to determine the actual size of the LOB segments and how to free the deleted/unused space above/below the HWM
Note 563470.1 – Lob retention not changing when undo_retention is changed
Note 422826.1 – How to identify LOB Segment Use PCTVERSION or RETENTION from Data Dictionary
Error Tracing

Undo error tracing can be done for normal undo operations using the following events:

NOTE: Normal undo operations will be indicated in the error message in that the error message includes a segment name like

…. name "_SYSSMU1$" too small

If the error doesn’t show a segment name

… name "" too small

the problem is often related to LOB undo
If using pfile:

event="10442 trace name context forever, level 10"

If using spfile:

Alter system set events '10442 trace name context forever, level 10';

Reproduce the ORA-1555 error and upload the trace file to Oracle Support.

LOB undo error tracing is more difficult. Set additional tracing events as follows:

Start Session 1
Alter session set events '10046 trace name context forever, level 12';
Reproduce the error
Exit Session 1

Start Session 2
Alter session set events '10051 trace name context forever, level 1';
Reproduce the error
Exit Session 2

Start Session
Alter session set events '1555 trace name errorstack forever, level 3';
Reproduce the error
Exit Session 3

Additional resources to review:
Note 846079.1 – LOBs and ORA-1555 troubleshooting
Note 253131.1 –Concurrent Writes May Corrupt LOB Segment When Using Auto Segment Space Management
Note 467872.1 – TROUBLESHOOTING GUIDE (TSG) – ORA-1555
V$UNDOSTAT Analysis

The V$UNDOSTAT view holds undo statistics for 10 minute intervals. This view
represents statistics across instances, thus each begin time, end time, and
statistics value will be a unique interval per instance.

This does not track undo related to LOB
Note 262066.1 – How To Size UNDO Tablespace For Automatic Undo Management
Note 1112363.1 – When Does Undo Used Space Become Available?
Note 240746.1 – 10g NEW FEATURE on AUTOMATIC UNDO RETENTION
Diagnostics Scripts

Refer to Note 746173.1 : Common Diagnostic Scripts for AUM problems
and Note 877613.1 : AUM Common Analysis/Diagnostic Scripts
Common Causes/Solutions

Using Rollback Segments functionality:

* Problem happening on SYSTEM tablespace that still uses old Rollback Segment functionality even when configured for Automatic Undo Management (AUM).

* There are not enough rollback segments to manage the undo needed for long running queries.

* Rollback Segments are too small and undo is overwritten before long running queries complete.
Reference:

Note 69464.1 – Rollback Segment Configuration & Tips
Note 10630.1 – ORA-1555: “Snapshot too old” – Overview
Note 862469.1 – ORA-604 & ORA-1555 Rollback Segment 0 with Name “System” Too Small
Using Automatic Undo Management (AUM):

* TUNED_UNDORETENTION in V$UNDOSTAT around the time of the error is lower than the QUERY DURATION indicated in the error message. This is a legitimate ORA-1555 and if queries are going to run for very long time frames, UNDO_RETENTION may need to be larger. Auto-tuned retention may not be able to keep up with the undo workload and staying within space limitations on the UNDO tablespace.

* LOB updates and/or deletes are frequent and a higher PCT_VERSION is required to provide enough space in the LOB Segment to accommodate the LOB undo. RETENTION on LOBs that are updated or deleted frequently can run into problems holding UNDO long enough for queries.

* QUERY DURATION shown in the error message is 30+ years and therefore, no amount of undo will satisfy the consistent read blocks.

Note 750195.1 – ORA-1555 Shows Unrealistic Query Duration (billions of seconds)

* QUERY DURATION shown in the error message is 0. NOTE: This has been filed as a bug on many release levels and has been very difficult to narrow down to a specific problem.

Note 761128.1 – ORA-1555 Error when Query Duration as 0 Seconds

* QUERY DURATION is lower than TUNED_UNDRETENTION. Undo header information can sometimes get overwritten or you could be seeing a bug.

* TUNED_UNDORETENTION stays very high and UNDO tablepsace continues to grow continuously or getting space errors.

Note 1112431.1 – Undo Remains Unexpired When Using Non-autoextensible Datafiles for Undo Tablespace.

Reference metalink Doc ID 1307334.1

No comments:

Post a Comment