[Unison-hackers] Memory exhaustion issue (#1068)

[Michael von Glasow]

I just attempted to set up a test environment to reproduce my
observations. The test system is a VirtualBox VM running Ubuntu Server
24.04.1, with 1 GB memory (961.7 MB as reported by top). Swap was
disabled for the test. Unison version is 2.53.3 (ocaml 4.14.1), from the
Ubuntu repositories.

I wrote a script to generate a directory of 700,000 files on my local
hard disk, each containing 1 kB of random data. Then I attempted to sync
that to the test machine, while running `top` to monitor resource usage.
At that time, around 430 MB of memory were used (not counting
buffers/cache) on the VM.

During most of the initial scan, Unison on the server was at 8.1 MB res;
when the scan finished, it was at 8.5 MB.

As I initiated sync, memory went up. (File counts are from the file
displayed by the Unison GUI; files are numbered 00001 to 70000 and were
created in that order.)

At 35000 files: 33.5M
At 70000 files: 51.6M
At 140000 files: 89.7M
At 210000 files: 126.0M
At 280000 files: 170.1M
At 350000 files: 213.0M
At 420000 files: 250.1M
At 490000 files: 290.3M
At 560000 files: 339.1M
At 630000 files: 377.3M
At 686000 files: 478M
At 700000 files: 523.5M

That is, initially memory consumption increases fairly constantly with
the number of files, about 40 MB per 70,000 files, which would be some
570 bytes per file. After reaching 490,000 files (70%), this goes up
slightly (the next 70,000 files occupy about 50 MB), then returns to the
old rate until, above 630,000 files, memory usage increases
dramatically: the next 56,000 files occupy around 100 MB of memory,
which would correspond to some 1.8 kB per file, almost twice the size of
a file. That is, unless one of these figures lags behind the other. On
average, one file occupies around 730 bytes of memory.

Unfortunately, sync did not finish but aborted with an error message
after (reportedly) transferring file 700,000:

Fatal error – Error in Lock.acquire: No space left on device

The path to the lock file is the one on the server (the hex value after
`lk` matches up with the `ar and `fp` files).

The server disk has some 4 GB of free space, so it looks like I hit some
other limit, not sure which. Filesystem is ext4, same as on the host system.

Files on the server seem to have been synchronized until 659751. The ar*
file is 655 bytes in size after the failed sync, the fp* file is 34
bytes in size. Memory usage of the Unison process on the server was at
524.2 MB, meaning we were close to running out of memory.

If I retry sync (with a fresh Unison process on both ends), I get the
same error message right after selecting the profile.