Upload
mongodb
View
1.152
Download
0
Tags:
Embed Size (px)
DESCRIPTION
MongoDB Management Service (MMS) is is a cloud-based suite of services for managing MongoDB deployments, providing both monitoring and backup capabilities. In this webinar we'll outline 5 alerts you should set up in MMS to keep your MongoDB deployment on track. We’ll explore what each alert means for a MongoDB instance, as well as how to calibrate the alert triggers to be relevant to your environment.
Citation preview
Five MMS Monitoring Alerts to Keep Your MongoDB Deployment on Track
Angshuman Bagchi ([email protected])Technical Services Engineer
Agenda
• What is MMS Monitoring?• What are Alerts?• How to pick an Alert?• Five recommended Alerts• Wrap up
What is MMS Monitoring?
Who uses MMS?
What are MMS alerts?
Source:http://www.cleanfunnypics.com/no-its-not-empty/#axzz2pqknJJbC
How to pick an Alert?
• Is there an absolute limit to alert on?• What is normal (baseline) ?• What is worrying (warning) ?• What is a definite problem (critical) ?• Likelihood of false positives ?
... there is no magic formula
Five recommended alerts
• Host Recovering (All, but by definition Secondary)
• Replication Lag (Secondary)• Connections (All mongos, mongod)• Lock % (Primary, Secondary)• Replica (Primary, Secondary)
Host Recovering
• General alert triggered if any instance enters RECOVERING mode
• Required for all use-cases• All Replica Sets should have this. • Sometimes, during maintenance this
may be expected
Host Recovering
Replication Lag
• No secondary should be behind• Secondary reads affected• All Replica Sets should have this• Only exception is configured slaveDelay
Replication Lag
Absolute Limit?Yes, about 1 or 2s. To prevent false positives absolute threshold > 240s should be alerted
Normal Lag is ideally 0s
Worrying < 60s, some false positives
Critical > 240s
False positives Above 240s likelihood low.
Example: replication lag
150,000s of lag ~ almost 2 days of lag!
Example: replication lag
• Secondaries under specified vs primaries• Access patterns between primary /
secondaries• Insufficient bandwidth• Foreground index builds on secondaries
“…when you have eliminated the impossible, whatever remains, however improbable, must be the truth…” -- Sherlock Holmes
Sir Arthur Conan Doyle, The Sign of the Four
Example: replication lag
Example:• ~1500 ops per minute (opcounters)• 0.1 MB per object (average object size,
local db)
~1500 ops/min / 60 seconds * 0.1 MB/op * 8b/B =~ 20 mbps required bandwidth
Connections
• Each connection consumes ~ 1MB and a file descriptor
• 5000 connections => 5GB of RAM• Stability and predictability are key
Pro-Tip: know thyself
You have to recognize normal to know when it isn’t.
Source: http://www.flickr.com/photos/skippy/6853920/
Connections
Absolute Limit? Yes, but this is too high. We need to alert before that
NormalTBD based on deployment, number of nodes, connection pool settings, app servers, load etc. Say, X during peak load
Worrying 50% increase, so, 1.5X
Critical Double, so 2X
Lock %
• Lock contention degrades performance• High lock % starves replication, reads.• Bounds need to be determined
Lock %
Absolute Limit?Yes, >80% occasional degraded performance, 90% major impact regularly
NormalTBD. Write heavy loads see higher values. Normal, say X% during peak load
Worrying Double, so approximately 2X%
Critical TBD. For Prod > 80%
Replica
• Represents oplog window• Depends on
– Rate of operations inserted into oplog– Size of operations– Size of oplog capped collection
• Normal maintenance window X 3 • Resizing the oplog is non-trivial
Replica
Absolute Limit? 50% below Normal
Normal TBD. Say X hours during peak
Worrying 25% below Normal
Critical 50% below Normal
Summary
• Use similar approach for other metrics• Different audiences for alerts
– Worrying alerts ops team– Critical goes out to a wider audience
• Get started with MMS Monitoring and alerts!
I got alerted … now what?
mms.mongodb.com