Does anyone have experience running Vertica in privileged Docker containers? Are there any significant performance implications or other pitfalls I should be aware of?
I can't speak to customer deployments; I would be very curious to hear from customers who have real field experience doing this.
One subtle pitfall to be aware of: Vertica assumes that it has access to all physical memory on your system, up to whatever you configure it to use. "Physical memory" is not the same as "whatever cap you place on your container"; Docker doesn't (to my knowledge) spoof the machine's listed physical memory the way VMs do, and Vertica doesn't check for container limits. So you need to update Vertica's resource pools to reflect any RAM cap that you impose on the container; otherwise Vertica will exceed the container's memory limits and be killed by the Linux kernel.
There are also all the usual pitfalls of running Vertica in a shared environment: If some other process slows down the machine, it'll make your whole Vertica cluster slow; etc.
Comments
One subtle pitfall to be aware of: Vertica assumes that it has access to all physical memory on your system, up to whatever you configure it to use. "Physical memory" is not the same as "whatever cap you place on your container"; Docker doesn't (to my knowledge) spoof the machine's listed physical memory the way VMs do, and Vertica doesn't check for container limits. So you need to update Vertica's resource pools to reflect any RAM cap that you impose on the container; otherwise Vertica will exceed the container's memory limits and be killed by the Linux kernel.
There are also all the usual pitfalls of running Vertica in a shared environment: If some other process slows down the machine, it'll make your whole Vertica cluster slow; etc.
Adam