Category: Administration

Laurenz Albe shares a warning:

Linux tries to conserve memory resources. When you request a chunk of memory from the kernel, Linux does not immediately reserve that memory for you. All you get is a pointer and the promise that you can use the memory at the destination. The kernel allocates the memory only when you actually use it. That way, if you request 1MB of memory, but use only half of it, the other half is never allocated and is available for other processes (or the kernel page cache).

Overbooking is a concept that airlines all over the world have been using for a long time. They sell more seats than are actually in the plane. From experience, airlines know that some passengers don’t show up for the flight, so overbooking allows them to make more profit. By default, Linux does the same: it deals out (“commits”) more memory than is actually available in the machine, in the hope that not all processes will use all the memory they allocate. This memory overcommit is great for using resources efficiently, but it has one problem: what if all the flight passengers show up, that is, what if the processes actually use more memory than is available? After all, you cannot offer a computer process a refund or a free overnight hotel room.

Read on to learn more about memory overcommit and what you can do about it.

Comments closed

Andy Brownsword lets things get out of hand:

Last week we looked at using Database Snapshots to help with rolling back upgrades. The snapshot maintained a point in time copy of the database which could be later restored.

We can go further – a database can have multiple snapshots.

Let’s suppose we want to take one before an upgrade, another once the upgrade is complete, and another before the start of business the following day. This would provide us multiple points to restore too.

This however makes restoring more complicated.

My recollection is that it’s not just restoration that gets more complicated, but also any database activity, to the point where too many database snapshots on a single database can have a considerable performance impact.

Comments closed

David Wheeler talks pre-loading:

Recently I’ve been trying to figure out when a Postgres extension shared libraries should be preloaded. By “shared libraries” I mean libraries provided or used by Postgres extensions, whether LOADable libraries or CREATE EXTENSION libraries written in C or pgrx. By “preloaded” I mean under what conditions should they be added to one of the Shared Library Preloading variables, especially shared_preload_libraries.

The answer, it turns out, comes very much down to the extension type. Read on for details.

Read on for an interesting discussion of what pre-loading means and the circumstances you should consider along the way.

Comments closed