Skip to main content

Proactive production support by trusting data

Five years ago when I joined my employer we were in reactive mode when it comes to production performance. Nodes were going down daily and we would jump to fix that fire and then move to next fire. Problem with this approach was that we would be always busy and even nightly calls for customers in Europe.  Similar pattern was happening when it came to production support, many issues were known only when a customer would report it. Again this would mean that we would get calls on weekend and we had to fix issues ASAP and sometimes that means putting a band-aid. Some times issues will linger in production unnoticed for 1-2 weeks and when customer would report it, its a lengthy and arcane process to hunt for issues in existing logs. More than debugging this would require  a lot of patience and persistence to look for trend and form a time-line of events to hunt for root cause. Only few engineers would be willing to do so.

Over the course of last few years we became better at production support by turning to proactive production support. I wrote a exception analysis report and  daily my team members and me would analyse exception report and send tickets to appropriate colleagues.  Many times there will be tickets that don’t require patch and can be temporarily fixed by doing some data cleanup. If the issue was in the flow executed by background jobs or automated agents then we would fix issues even before customer would notice it.  Advantage of this approach is that issues will get caught before it becomes a nightmare. Incremental fixes adds up to a lot and now the volume of production issues we have to fix has reduced because we fix many of them in the background and include it in coming patch.

Two months ago we installed new relic and we are not turning the table around on performance front also by trusting data rather than gut feelings.  New relic allows us to do trend analysis and release impact analysis, we can catch the issues while its turning from Green->Yellow->Red. Off course some issues turn from Green to Red within 10-15 mins but there are many that takes time. For e.g. Last week I noticed that one data centre is behaving odd and avg response time of all apis increased from 50ms to 100ms. That means customers were having degraded performance but still they were not complaining. Today I found it was due to one customer and the problem can be fixed by some data tweaks. I did the tweaks and performance is back to normal.

Similarly from new relic we found many issues where calls to memcached would happen in a loop or database queries in a loop. Those kind of issues affect customers with 25-30K users but not all customers. Every penny counts and by taking a proactive jab at these issues we are increasing customer satisfaction and loyalty but also we are becoming more predictable by trusting metrics.

for e.g. we recently converted a service from python to java and in past we have never ramped up from 0 to 1000 customers in a week.  But because of new relic I clearly saw that system was performing well with no errors even though we ramped from 30->200->1000 clients. Tomorrow we would add 500 more clients and all this is because the decision is based on data rather than gut feeling.



Popular posts from this blog

RabbitMQ java clients for beginners

Here is a sample of a consumer and producer example for RabbitMQ. The steps are
Download ErlangDownload Rabbit MQ ServerDownload Rabbit MQ Java client jarsCompile and run the below two class and you are done.
This sample create a Durable Exchange, Queue and a Message. You will have to start the consumer first before you start the for the first time.

For more information on AMQP, Exchanges, Queues, read this excellent tutorial
import com.rabbitmq.client.Connection; import com.rabbitmq.client.Channel; import com.rabbitmq.client.*; public class RabbitMQProducer { public static void main(String []args) throws Exception { ConnectionFactory factory = new ConnectionFactory(); factory.setUsername("guest"); factory.setPassword("guest"); factory.setVirtualHost("/"); factory.setHost(""); factory.setPort(5672); Conne…

Logging to Graphite monitoring tool from java

We use Graphite as a tool for monitoring some stats and watch trends. A requirement is to monitor impact of new releases as build is deployed to app nodes to see if things like
1) Has the memcache usage increased.
2) Has the no of Java exceptions went up.
3) Is the app using more tomcat threads.
Here is a screenshot

We changed the installer to log a deploy event when a new build is deployed. I wrote a simple spring bean to log graphite events using java. Logging to graphite is easy, all you need to do is open a socket and send lines of events.
import org.slf4j.Logger;import org.slf4j.LoggerFactory; import; import; import; import java.util.HashMap; import java.util.Map; public class GraphiteLogger { private static final Logger logger = LoggerFactory.getLogger(GraphiteLogger.class); private String graphiteHost; private int graphitePort; public String getGraphiteHost() { return graphiteHost; } public void setGraphite…

What a rocky start to labor day weekend

Woke up by earthquake at 7:00 AM in morning and then couldn't get to sleep. I took a bath, made my tea and started checking emails and saw that after last night deployment three storage node out of 100s of nodes were running into Full GC. What was special about the 3 nodes was that each one was in a different Data centre but it was named same app02.  This got me curious I asked the node to be taken out of rotation and take a heap dump.  Yesterday night a new release has happened and I had upgraded spymemcached library version as new relic now natively supports instrumentation on it so it was a suspect. And the hunch was a bullseye, the heap dump clearly showed it taking 1.3G and full GCs were taking 6 sec but not claiming anything.

I have a quartz job in each jvm that takes a thread dump every 5 minutes and saves last 300 of them, checking few of them quickly showed a common thread among all 3 data centres. It seems there was a long running job that was trying to replicate pending…