RE: Log in Storm Caused Database Crash
Date: Mon, 2 Oct 2017 09:12:15 -0400
Message-ID: <02e201d33b80$12e27500$38a75f00$_at_rsiz.com>
A typical (not to be confused with good) approach from application servers is that if a connection is not available, some number of connections for a pool is spawned.
Too often this approach is implemented without any semaphore observable from the application server to prevent immediately succeeding connection requests from driving another pool spawn and instead to retry first after the already requested spawn has completed.
This pattern (“I didn’t get a connection, I’ll help everyone else by getting a connection for myself and also for the next, say, 60, attempters, so they don’t have to wait”) is a simple minded recipe for disaster waiting to happen when a connection request burst (whether denial of service, load testing, or simply clock driven like 9 AM in an office setting) fields requests at a rate quicker than the time to make the additional connections available.
The load on the system from a single request in resource and lag time goes up by a factor and since it cannot service quick successive requests, these each in turn add to the problem.
This may or may not be your problem, but it is worth checking for.
mwf
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Ravi Teja Bellamkonda
Sent: Monday, October 02, 2017 1:06 AM
Matthew and Stefan,
Thank you for your points. I will definitely go through those logs.
During that interval, I made sure from v$session (machine) that all the sessions are from the application server itself. This made me feel that the spike is because of the application server configuration during load times.
On Sun, Oct 1, 2017 at 9:58 PM, Matthew Parker <dimensional.dba_at_comcast.net> wrote:
You can also look at you listener log to see number of connection attempts to system.
If it was a true overwhelming connection storm then you will also probably have information in the alert.log where errors are being reported spawning against the database.
Matthew Parker
Chief Technologist
Dimensional DBA
425-891-7934 <tel:(425)%20891-7934> (cell)
D&B 047931344
CAGE 7J5S7
<http://www.linkedin.com/pub/matthew-parker/6/51b/944/> View Matthew Parker's profile on LinkedIn
<http://www.dimensionaldba.com/> www.dimensionaldba.com
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Ravi Teja Bellamkonda
Sent: Sunday, October 1, 2017 9:45 PM
Hi Upendra,
Thank you for your response. This is an internet facing application and we were expecting a burst load to check for the capacity of the system. Is there a way to measure what no of sessions in the database is breaking point. I was doubting if any Sub-Optimal Connection Pooling might have caused this.
Highly appreciated your help here.
On Sun, Oct 1, 2017 at 7:39 PM, Upendra nerilla <nupendra_at_hotmail.com> wrote:
Is this an internet facing application or internal? If it is external facing application, investigate if there was DoS type attack or a spike in the user sessions due to any issues with application servers?
If you need to isolate where the connections originated from, you could look into DBA_Hist views.
You may want to start with this one.. https://docs.oracle.com/cd/B19306_01/server.102/b14237/statviews_3125.htm#REFRN23400
<https://docs.oracle.com/cd/B19306_01/server.102/b14237/statviews_3125.htm#REFRN23400> DBA_HIST_ACTIVE_SESS_HISTORY - Oracle Help Center
docs.oracle.com
DBA_HIST_ACTIVE_SESS_HISTORY. DBA_HIST_ACTIVE_SESS_HISTORY displays the history of the contents of the in-memory active session history of recent system activity.
Also look into any application server logs and see if there were any issues with the application server itself..
-Upendra
From: oracle-l-bounce_at_freelists.org <oracle-l-bounce_at_freelists.org> on behalf of Ravi Teja Bellamkonda <raviteja.bellamkonda7_at_gmail.com>
Sent: Sunday, October 1, 2017 9:25 PM
Hi List,
We ran into an issue recently and wanted some help in figuring out this issue.
Database was not responding and one thing from AWR observed before fail over was the login storm.
Inline image 1
Logons cumulative also increased during this interval.
To: Matthew Parker
Cc: Upendra nerilla; oracle-l
Subject: Re: Log in Storm Caused Database Crash
<mailto:Dimensional.dba_at_comcast.net> Dimensional.dba_at_comcast.net
To: Upendra nerilla <nupendra_at_hotmail.com>
Cc: oracle-l <oracle-l_at_freelists.org>
Subject: Re: Log in Storm Caused Database Crash
To: oracle-l
Subject: RE: Log in Storm Caused Database Crash
Inline image 2
Logons cumulative were 1237 in total in the before AWR report. Any suggestions are highly appreciated.
-- Thanks & Regards, Ravi Teja -- Thanks & Regards, Ravi Teja Bellamkonda -- Thanks & Regards, Ravi Teja Bellamkonda Ph: (816)-905-7577.Received on Mon Oct 02 2017 - 15:12:15 CEST
-- http://www.freelists.org/webpage/oracle-l
- image/png attachment: image001.png
- image/png attachment: image002.png