RE: combining metrics in oem 12c
Date: Fri, 13 Feb 2015 06:48:08 -0800 (PST)
Message-ID: <f14b6518-4511-4008-99b7-751d5a41f6f1_at_default>
If you want warnings “in bulk” you can use the reports to send an email of all warnings at X hr. The notification system is meant for instant notifications.
From: Ethan Post [mailto:post.ethan_at_gmail.com]
Sent: Friday, February 13, 2015 8:41 AM
To: Ls Cheng
Cc: Oracle Mailinglist
Subject: Re: combining metrics in oem 12c
One thing I have found helpful is using a buffering system, don't know if this can be implemented in EM but worth checking. This groups emails together into a single message and sends every N minutes where N is usually 5 for a production system and as long as 2-4 hours for non-production. These are warnings, a critical event such as space issue, service outage, sends a message (email/sms) right away and causes a immediate dump of all buffered emails. Also labeling helps identify production from non-production very quickly. In your example that might be what I consider a warning, although I would never warn on that actually.
Question for the EM experts, can you share monitoring configurations easily? Is this just an XML file or something? If so has anyone published an fairly good config?
On Fri, Feb 13, 2015 at 3:17 AM, Ls Cheng <HYPERLINK "mailto:exriscer_at_gmail.com" \nexriscer_at_gmail.com> wrote:
Oh, no that was a example which took me 15 second to write it down, it does not make much sense combining logical reads and CPU as you have suggested.
The problem I have now is that I have database adaptive threshold on and it sends quite a lot of mails (may be I set too many metrics) and most notification are really critical. For example it mails to notify one of the databases is having 5500 iops, ideally it should notify if the access time is also high.
I will look into incident and notification and see if it is doable cmbining metrics
Thanks
On Fri, Feb 13, 2015 at 5:57 AM, Ethan Post <HYPERLINK "mailto:post.ethan_at_gmail.com" \npost.ethan_at_gmail.com> wrote:
No idea but likely you can write any custom type alert you desire in 12c. I have written my own system over the years for this sort of thing I can tell you I have never seen a circumstance in which I want to get out of bed for such a thing. Don't know what issue your trying to critical alert on but I am sure I can come up with a better scenario which won't capture things you don't care about.
Why logical reads? Why not CPU? For example, if more than 5 sessions have used more than 75% of a single CPU within the last hour and total available CPU is < 20% for last 10 minutes, then critical event. Or better yet, send emails until 6AM, then start paging/open ticket. This usually allows you to get some much needed sleep but get alerted before most of the folks roll into work. Or just if total CPU is <20% available for past hour minus the CPU datapump is burning. You want to alert on the exact issue you are getting dinged for and still, only get out of bed if it can't wait until 6 or 7 am. Trust me on this, sleep is awesome and makes for happy employees.
I have stayed away from EM because by the time I figured out how to write such an alert in EM terms I usually start wishing I wrote it from scratch. I know EM has improved a lot so not going to knock it, I am just out of the loop totally on it, but logical IO is just not a good starting point for an alert usually, at least one that is going to port well to other databases (like when CPU's get faster).
On Thu, Feb 12, 2015 at 5:00 PM, Ls Cheng <HYPERLINK "mailto:exriscer_at_gmail.com" \nexriscer_at_gmail.com> wrote:
Hi
Does anyone know if it is possible in Enterprise Manager 12c R4 combine metrics to produce notifications?
For example if session logical reads metric is 100000 per second and average active session is larger than 5 then generate a critical event?
TIA
-- http://www.freelists.org/webpage/oracle-lReceived on Fri Feb 13 2015 - 15:48:08 CET