CodeParkBGSAVE calls fork() to create a child process. The child sees memory frozen at that moment and writes it to the RDB file. Meanwhile, the parent keeps serving commands.
This works because of copy-on-write. The parent and child share the same memory pages. When the parent changes a page, the operating system copies that page first. Only changed pages are copied.
The risks:
- Many writes during the snapshot change many pages. In the worst case, memory use gets close to double the dataset.
- The
fork()call itself can pause the main thread briefly on very large instances, because the page tables must be copied.
What to do:
- Leave free memory below the machine's RAM. Do not set
maxmemoryclose to it. - Turn off Linux transparent huge pages for Redis. They make each page much bigger, so every copied page costs much more memory.
fork()creates a child process that writes the snapshot.- Copy-on-write shares memory pages until the parent changes them.
- Heavy writes during a save can nearly double memory use.
- The fork itself can cause a short latency spike on big instances.
- Leave free memory headroom and turn off transparent huge pages.