Rarely you might have heard of people running database in container in production environment. If there are any, gutsy they must be.
Database is critical component of every three tier service. With advent of NoSQL especially Document type databases, like MongoDB, database selection also play a role in deciding your technology stack. Though with failure of any component in stack may take down your whole service along, however, with database down you also looses data.
Critics may say, Hey, same is true for any stateful component. But how many stateful component you see beside database nowadays? Smart people don’t uses file for persisting any information. Almost, all application states including sessions are stored in persistent fast accessible layer which will be 99% a database.
So, Should i dare to run database in Docker?
Very daunting question it might look but if we paraphrase it, it looks like this:
With database running,
- I should not lost my data.
My data should persist on disk and should stay there even when service is stop.
- I should be able to collect database metrics
Most of the metrics usually come from logs. Need not to mention logs also help us in debugging in case some anomally occur in operation.
- I should be able to shutdwon gracefully, if not in use
Some database require cleaning operation on shutdown phase. It also remove lock and release resources in cleaning step. When database is forcefully killed some resources not get released. Usually database respond to SYSINT and SYSKILL command by capturing them and taking decision.
- I should be able to start it with last saved data.
When database re-start, it should resume operation from last saved state. State is usually saved in files(disk). So, effective journaling should be there.
- I should have no latency i.e
- no network latency
In distributed environment network latency can become a overhead. When there are thousands of IOPS per seconds it can easily become bottleneck.
no disk i/o latency
Though memory and journaling help in minimizing disk i/o operations. But a faster disk access always lead to faster throughput. With hardwares getting cheaper, SSD has gained popularity. Being fast often means fast operation.
I should be able to scale it
Through master-master replica or master-slave replica
Not that daunting now. Right?
Let me recap docker performance impact on application running inside it. In performance, it’s tantamount to speed you get natively. Here’s a brief overview on it.
CPU : CPU performance is totally Native. Same as you get without running inside Docker.
I/O : Docker support many storage backends. Device Mapper is being used by default and can be switched as per need. However, being said this simply switching storage backend won’t get you faster iops that you get natively. However, if you use docker volumes you’ll get native speed. It also has few extra advantages which i’ll cover in some future post.
Memory : Docker set aside a little memory for memory accounting. However, it can be native i.e No overhead, if you disable memory accounting (useful for hptc, probably not for everything else)
Network: There is no overhead if you run with
-–net host. It’s useful for > 1Gb/s workloads of if you have a high packet rate eg. VOIP, Gaming.
I might be missing few other things connected especially to application like Database. Pardon my ignorance. Here, i’d appreciate few help from community to add their views in the comment.