![]() Minutes of coding, and dozen more minutes for playing with Write some ugly python code I won’t dare to show here. To parse the dates from logs and convert them to unix time, I had to Line, the timestamp and the latency (or whatever you want to map): $ tail -n 5 ~/tmp/responses.dat The input format is simple: just two space-separated numbers per Sources like perf or iosnoop, it’s in fact quite generic. While this tool is usually associated with some hardcore data To build heat maps from traces gathered with various tools, most We’ll be able to see how response time changes as JVM and Solr cachesīrendan Gregg once created a very simple, easy to use perl script Let’s extract the repsonse times and build some heatmaps! For theĮxperiment, I’m taking response times for the few minutes after theĪpplication startup, with all the cache warming disabled. Relevant message may look like this: 13:42:33,559 INFO - Client request: /search?keyword=denim%20shirt, response time: 57 ms, found=73, stage=1 ![]() Testbeds, or production – in form of the application logs, where a ![]() I canĮasiliy gather the response times from the service – from various The Experiment ¶Ĭurrently, I work on a Solr-based search service for one of the USĮ-retailers, and of course I’m interested in its performance. How to draw them? Below, I’ll take a quick look onto a couple of ways. Heatmaps are awesome: in many cases, using color to represent an extraĭimension allows to get good insights on what’s really happening. ![]()
0 Comments
Leave a Reply. |