Troubleshooting - Practical Cassandra - page 159

24753.60 122.40 67.40 92.58

await svctm %util

2976.68 2.39 88.00

Again, it is important to note that iostat is an indicator of a potential prob-
lem, not an absolute. So take the results of analyzing the output of iostat into
account when looking into issues, but don't use them as an absolute measure of a
problem. One of the primary assumptions being made when asking the question
“How much time does it take Cassandra to read from the disk?” is that Cassandra
is the primary application on the machine competing for disk access.

dstat

dstat is another excellent tool for seeing the overall health of your system. Like
iostat , dstat does not come installed with most common operating systems.
Listing 10.3 shows an example of dstat output on a healthy active Cassandra
cluster. There are a few switches that are handy to be aware of. The -l switch tells
dstat to print the system load information. The -n switch tells dstat to print
the network information. The -v switch tells dstat to show the output in a vm-
stat style ( vmstat is short for Virtual Memory Statistics, which is another *nix
tool). The -r switch tells dstat to show I/O request information. The 10 tells
dstat to show a new line once every ten seconds.
Listing 10.3 Example Output of dstat on a Healthy Active Cassandra
Cluster
Click here to view code image

# dstat -lnvr 10
...load-avg... -net/total- ...procs...
......memory-usage..... -paging-
1m 5m 15m| recv send|run blk new| used

buff cach free| in out |
1.15 1.34 1.57| 0 0 | 0 0.0 1.8|10.3G 3164k

4421M 71.5M| 0 0 |
1.35 1.38 1.58|1197k 863k|2.0 1.6 0.4|10.3G 3172k
4417M 75.1M| 0 0 |
1.22 1.35 1.57|1230k 936k|1.1 1.0 1.9|10.3G 3172k
4422M 70.8M| 0

0 |