1

I have a large data set, and just enough RAM to store most of my data (I have some unique access patterns). I want to see if my DB performance is degrading because my working set is too large for my hardware.

How would I got about doing this? Which Linux tools can I use. I am currently using Mongo 2.2 on Ubuntu 11.04. Which metrics should I be looking at?

Update: I guess a better way to ask this question would be "What tools to experienced mongo admins use to monitor performance?"

josephmisiti
  • 9,862
  • 11
  • 56
  • 72
  • Did you try first to use `top` ? – Basile Starynkevitch Jan 23 '13 at 06:45
  • http://www.10gen.com/products/mongodb-monitoring-service and also please take a look at the following answer of the question: http://stackoverflow.com/questions/14061833/mongodb-count-with-queries-have-poor-performance – cubbuk Jan 23 '13 at 07:38
  • 1
    The best of the best is MMS plus you can get proper support from 10gen through it since they can see the stats, I would go for MMS, its a no brainer. Also unlike what the linked answer says, INDEXES DO NOT NEED TO FIT INTO RAM, your working set does. That is a common misconception – Sammaye Jan 23 '13 at 08:38
  • 1
    if you just want to go for some quick check you can also use mongostat and mongotop. See: http://docs.mongodb.org/manual/reference/mongostat/ and http://docs.mongodb.org/manual/reference/mongotop/ – philnate Jan 23 '13 at 09:31
  • i quit using top after I read this:http://docs.mongodb.org/manual/faq/diagnostics/#how-do-i-read-memory-statistics-in-the-unix-top-command – josephmisiti Jan 23 '13 at 16:08

3 Answers3

1

Mongostat should give you most of the information you're looking for, as far as the performance goes. The key metrics to look at would be index misses, page faults, lock percent, and the queue lengths for readers and writers. If these numbers are high (anything above zero, ideally, more so if they are not changing much) then you've a problem.

As far as tools go, MMS is good enough to start, but personally I've found myself resorting to mongostat every time I had to look into performance. Also, look at the logs, based on the profile level you should be seeing slow queries in there, which should give you an idea of which queries (or operations) to optimize.

lobster1234
  • 7,679
  • 26
  • 30
0

I found some Monitoring commands here. Could this be what you are looking for? As for the tools that you would use to to track the problem, here s how I do it.

Nagios: I use nagios as a centralized monitoring system to check the health of each servers. It alerts me if somethings goes wrong. For example, I have alerts set for situations like if my response time to a API goes beyond a threshold value.

Monit: If something goes totally crazy, monit takes care of it. Suppose any component of my stack goes down. It alerts me and also bring it up for me.

Logstash: Any suspicious activity recorded in my logs, it lets me know.

SeaLion: Now all the above lets me know if there was a problem with my stack. But what caused the problem, I debug with SeaLion. I log the outputs of system profiling tools like sar, top, uptime, iostat, vmstat, netstat etc. Besides these commands you can setup the commands that you would require on SeaLion and it executes all these commands and shows it in a nice timeline format. Also installation and setup is extremely easy.

Kevin
  • 566
  • 2
  • 5
  • 13
0

iostat will tell you how your disk subsystem is handling the load. I like iostat -x -k 1 for a time-series view.

leif
  • 1,987
  • 4
  • 19
  • 22