Carlos Sierra's Tools and Tips

Tools and Tips for Oracle Performance and SQL Tuning

eDB360 takes long to execute!

with 2 comments

eDB360 provides a lot of insight about an Oracle database. It executes thousands of SQL statements querying GV$ and DBA views from the Oracle’s data dictionary. It was designed to impose very little load on the system where it executes, thus it consumes only one session and avoids parallel execution. On a system where the state of historical metrics is “normal”, eDB360 may take one or two hours to execute. In the other hand, when the volume of historical metrics is abnormally “large”, then eDB360 may execute for many hours up to default threshold of 24, then quit. No worries here, it can be restarted pretty much where it left…

If you are considering executing eDB360 on one of your databases, before you do please execute first a light-weight “pre-check”, which is basically a small script that reports on the state of Active Session History (ASH) stored into the Automatic Workload Repository (AWR). Bare in mind that ASH and AWR are part of the Oracle Diagnostics Pack, so consider such “pre-check” only if you have a license for such Oracle Pack for the target database.

AWR ASH Pre-Check

You can execute stand-alone the awr_ash_pre_check.sql script available as free software in GitHub within edb360-master or cscripts-master. If downloading eDB360, you can find awr_ash_pre_check.sql under the edb360-master/sql directory, and if you download the cscripts set, then you can find awr_ash_pre_check.sql under cscripts-master/sql.

This pre-check script reports on the following potential concerns:

  1. Retention period of AHR, frequency of Snapshots, AWR Baselines, and similar.
  2. Age of CBO Statistics on the main table behind DBA_HIST_ACTIVE_SESS_HISTORY (WRH$_ACTIVE_SESSION_HISTORY) and its partitions.
  3. State of CBO Statistics for WRH$_ACTIVE_SESSION_HISTORY segments: Are the statistics Locked? Are they Stale?
  4. Size of WRH$_ACTIVE_SESSION_HISTORY segments, with range of Snapshots per segment (partition). Are they outside the desired retention window?
  5. Creation date and last DDL date for WRH$_ACTIVE_SESSION_HISTORY segments. So expected partition splitting can be verified.

Take for example result below. It shows that current ASH repository contains 320.6 days of history, even when in this case only 30 were requested (displayed on an upper part of report not shown here). We also see ASH is 16.4 GBs in size, where normally 1-2 GBs are plenty. We can conclude last partition split (or simply last DDL) is 289.7 days old, where normally only a few days-old are expected. This pre-check sample results on a ballpark estimate of 127 hours to execute eDB360. In other words, if we had one month of history (as current AWR configuration requests) instead of 320.6 days of history, then ASH repository would be less than 10% its current size and eDB360 would be estimated to execute in about 13 hours instead of 127. Keep in mind this “pre-check” provides a ballpark estimate, so an actual execution of eDB360 would take more or less the estimated time.

screen-shot-2016-11-23-at-2-03-55-pm

What to do if eDB36o is estimated to run for several days like in sample above? What I recommend is to fix ASH first. This process requires to read and digest at least 3 My Oracle Support (MOS) notes below, but it is not as bad as it sounds. You want to follow this path anyways, so any diagnostics tool that relies on ASH from AWR would benefit of the much needed cleanup.

  1. WRH$_ACTIVE_SESSION_HISTORY Does Not Get Purged Based Upon the Retention Policy (Doc ID 387914.1)
  2. Bug 14084247 – ORA-1555 or ORA-12571 Failed AWR purge can lead to continued SYSAUX space use (Doc ID 14084247.8)
  3. Manually Purge the Optimizer Statistics and AWR Snaphots to Reduce Space Usage of SYSAUX Tablespace (Doc ID 1965061.1)

Diagnosing eDB360 taking long

Let’s say you executed an older version of eDB360 and it took long (newer versions starting on v1620 perform an automatic pre-check, and if execution is estimated to last more than 8 hours, then it pauses and ask for confirmation before executing). If you executed an older version of eDB360, or simply your execution took longer than expected, review next the following files included on the directory from where you executed eDB360 (edb360-master), or stored within the edb360_*.zip file generated by eDB360 as it executed.

  1. awr_ash_pre_check_<database_name>.txt
  2. verify_stats_wr_sys_<database_name>.txt
  3. 00002_edb360_<NNNNNN>_log.txt
  4. 00003_edb360_<NNNNNN>_log2.txt
  5. 00004_edb360_<NNNNNN>_log3.txt
  6. 00005_edb360_157536_tkprof_sort.txt

If you find one of the SQL statements from eDB360 is taking more than a few minutes to execute, suspect first outdated CBO statistics on the Data Dictionary objects behind such query. Or if volume of data accessed by such query from eDB360 is large, suspect AWR data not getting purged. In either case take action accordingly.

If you cannot make sense of the information provided by these diagnostics files above, then contact eDB360 author at carlos.sierra.usa@gmail.com.

Keep in mind that execution time for eDB360 really depends on the state of several metrics stored in AWR, where ASH is the most relevant piece but not the only one. Think outdated statistics and outdated metrics!

Advertisements

Written by Carlos Sierra

November 23, 2016 at 5:47 pm

Posted in General

2 Responses

Subscribe to comments with RSS.

  1. How do you avoid the following prompt when trying to run edb360 in the background?

    *** eDB360 may take over 8 hours to execute, hit “return” to continue, or control-c to quit.

    Michael Fontana

    September 19, 2017 at 5:08 pm

    • Modify edb360-master/edb360.sql and set this:
      DEF ash_validation = ‘–skip–‘;

      Carlos Sierra

      September 19, 2017 at 5:36 pm


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: