Note: i never really considered putting oracle in kubernetes to be a technical problem; since you can pass a “raw” block device from a distributed block storage.<p>I consider it to be a licensing nightmare as the normal oracle license means you’ll need to license every single vCPU on the cluster.<p>The express edition used here has some relatively anaemic usage limits which you could brush up against very quickly:<p>“Oracle Database 18c Express Edition automatically constrains itself to the following resource restrictions;<p>2 CPU threads;
2 GB of RAM;and
12 GB of user data.”<p>Still, it’s cool, but I would caution anyone against <i>introducing</i> an oracle dependency.
If, like me, you at some point need to support Oracle coming from Postgres there are a few annoyances you should be aware of:<p>- No boolean type. Different folks will choose T/F chars, Y/N chars or 0/1 integers.
- Strings are not nullable. An empty string and a null string are the same thing.
- Date also has time.<p>By default, you end up with some very annoying bugs. Expect to actually change the way some of your columns work.
Oh my goodness, this is an awful idea. Oracle just runs hosted Exadata for OCI last I checked.<p>RAC requires shared block storage and L2 private network. Cloud SDNs and storage require gross hacks with horrible performance consequences. Even “modern” virtualization is painful, but can work.<p>Oracle isn’t going to fix all the bare-metal spaghetti assumptions in their clustered DB stack, and has been pretty clear about that.<p>This could be useful for throwaway dev/test environments. Or maybe for apps that aren’t performance intensive or critical that are in “maintenance mode” and folks want to lift-and-shift. But they’ll probably spend more time on that than fixing/replacing/retiring the app. And all will require smaller data sets than what I see with big company legacy systems.<p>And typically those systems are using Oracle to access data populated by another system, which makes me die a little inside.<p>Pretty clearly a play to get big companies into GCP contracts, more than anything real.<p>It’s to sell to execs who want to do the cool thing without paying to fix the old thing. <i>ahem</i> Thomas Kurian <i>cough</i>.
Would love to see Google throw this sort of effort at something other than Oracle. The Zalando crew have done an amazing job with Patroni and their Postgres operator. But having Google’s resources could really elevate in-cluster databases to the mainstream.