Why another article on this, Marco?

MySQL deadlocks is a topic covered many times, including here at Percona. I suggest you review the reference section at the end of this post for articles on how to identify deadlocks and from where they are generated.

So why another article?

The answer is that messages we receive like the following are still very common:

User (John): “Marco, our MySQL is having problems”
Marco: “Ok John what problems? Can you be a bit more specific?”
John: “Our log scraper is collecting that MySQL has a lot of errors”
Marco: “Ok can you share the MySQL log so I can review it?”
John: “Errors are in the application log, will share one application log”

Marco reviews the log and in it he finds:

“ERROR 1213 (40001): Deadlock found when trying to get lock;
try restarting transaction”

Marco’s reaction is: “Oh …

Oracle MySQL is in need of a columnar storage engine for analytics workloads.  A columnar engine (or column store) stores data vertically, that is, it stores all the data associated with a column together, instead of the traditional RDBMS storage method of storing entire rows together, either in a index organized manner, like InnoDB, or in a heap, like MyISAM.  

Columnar storage has the benefit of reducing IO when only a subset of the row is accessed in a query, because only the data for the accessed rows must be read from disk (or cache) instead of having to read entire rows.  Most columnar stores do not support indexes, but WARP does.

WARP is open source

You can find the WARP source code release on GitHub.  Binaries can be provided upon request.  Simply open an issue for your desired Linux distribution, and I will make them available as soon as I can.

…

Modern systems are complicated beasts with lots of interdependent activities between threads, programs and kernels. Figuring out some problems is nearly impossible without building some time machine and crystal ball mix that tells exactly what happened.

Did your cgroups CPU limit inject a sleep in the middle of mmap_sem acquisition by ‘ps’? Is everyone waiting for a mutex that is held by someone who is waiting for a DNS response? Did you forget to lock in your libnss.* libraries into memory and hence ended up stalling in unexpected place under memory pressure?

I’ve grabbed Brendan Gregg‘s offcpu profiler, gutted it to the point where all it does is record longest waits per stack trace, as well as timestamp of the longest wait.

Some debugging sessions that would’ve taken hours, days or weeks before now are few minute endeavors. It is still quite hacky, so …

Why another article on this Marco?

Deadlocks is a topic covered many times and with a lot of articles on the web, also from Percona.
I suggest you review the reference section for articles on how to identify Deadlocks and from where they are generated.
So why another article?
The answer is that messages like the following are still very common:

User (John): “Marco our MySQL is having problems”
Marco: “Ok John what problems. Can you be a bit more specific?”
John: “Our log scraper is collecting that MySQL has a lot of errors”
Marco: “Ok can you share the MySQL log so I can review it?”
John: “Errors are in the application log, will share one application log”

Marco reviews the log and in it he founds:

“ERROR 1213 (40001): Deadlock found when trying to get lock;
try restarting transaction”

Marco reaction is: "Oh my ..." 

…

Why another article on this Marco?

Deadlocks is a topic covered many times and with a lot of articles on the web, also from Percona.
I suggest you review the reference section for articles on how to identify Deadlocks and from where they are generated.
So why another article?
The answer is that messages like the following are still very common:

User (John): “Marco our MySQL is having problems”
Marco: “Ok John what problems. Can you be a bit more specific?”
John: “Our log scraper is collecting that MySQL has a lot of errors”
Marco: “Ok can you share the MySQL log so I can review it?”
John: “Errors are in the application log, will share one application log”

Marco reviews the log and in it he founds:

“ERROR 1213 (40001): Deadlock found when trying to get lock;
try restarting transaction”

Marco reaction is: "Oh my ..." 

…

Back to part 1

Now that we had seen how to setup our environment with PXC8 is time to see how our setup will behave and what can be done in case of issues.

We will now analyse the following common situations:
- Put a node in maintenance or remove a node and insert it back
- node crash
- network failure
- proxysql node crash

The following tests are done using a java application connecting using straight JDBC connection and no connection pooling. I choose to use that more than a …

Back to part 1

Now that we had seen how to setup our environment with PXC8 is time to see how our setup will behave and what can be done in case of issues.

We will now analyse the following common situations:
- Put a node in maintenance or remove a node and insert it back
- node crash
- network failure
- proxysql node crash

The following tests are done using a java application connecting using straight JDBC connection and no connection pooling. I choose to use that more than a …

In real life, there are frequent cases where getting a running application to work correctly is strongly dependent on consistent write/read operations. This is no issue when using a single data node as a provider, but it becomes more concerning and challenging when adding additional nodes for high availability and/or read scaling.

In the MySQL dimension, I have already described it here in my blog Dirty Reads in High Availability Solution.

We go from the most loosely-coupled database clusters with primary-replica async replication, to the fully tightly-coupled database clusters with NDB Cluster (MySQL/Oracle).

Adding components like ProxySQL to the architecture can, from one side, help in improving high availability, and from the other, it can amplify and randomize the negative effect of a …

In real life, there are frequent cases where getting a running application to work correctly is strongly dependent on consistent write/read operations. This is no issue when using a single data node as a provider, but it becomes more concerning and challenging when adding additional nodes for high availability and/or read scaling.

In the MySQL dimension, I have already described it here in my blog Dirty Reads in High Availability Solution.

We go from the most loosely-coupled database clusters with primary-replica async replication, to the fully tightly-coupled database clusters with NDB Cluster (MySQL/Oracle).

Adding components like ProxySQL to the architecture can, from one side, help in improving high availability, and from the other, it can amplify and randomize the negative effect of a …

The Percona Kubernetes Operator for Percona XtraDB Cluster(PXC) comes with ProxySQL as part of the deal. And to be honest, the behavior of ProxySQL is pretty much the same as in a regular non-k8s deployment of it. So why bother to write a blog about it? Because what happens around ProxySQL in the context of the operator is actually interesting.

ProxySQL is deployed on its own POD (that can be scaled as well as the PXC Pods can). Each ProxySQL Pod has its own ProxySQL Container and a sidecar container. If you are curious, you can find out which node holds the pod by running

kubectl describe pod cluster1-proxysql-0 | grep Node:
Node: ip-192-168-37-111.ec2.internal/192.168.37.111

Login into and ask for the running containers. You will see something like this:

[root@ip-192-168-37-111 ~]# docker ps | grep -i proxysql …