Use at most 8 threads for the xz stream

At preset 6, xz2 uses about 173MB of memory per thread. This adds up
quickly -- e.g. over 8GB of memory on a 48-CPU machine. If you happen to
try this in a 32-bit build, you'll get `LZMA_MEM_ERROR`.

We can limit this to a heuristic maximum number of threads to avoid
using so much memory, like xz's [`04_compress_easy_mt` example].

    // The number 8 is arbitrarily chosen and may be too low or
    // high depending on the compression preset and the computer
    // being used.

[`04_compress_easy_mt` example]: https://github.com/xz-mirror/xz/blob/de1f47b2b40e960b7bc3acba754f66dd19705921/doc/examples/04_compress_easy_mt.c#L71

1 files changed, 4 insertions, 2 deletions

diff --git a/src/tarballer.rs b/src/tarballer.rs
index 9ff1bd2..fa6ab2c 100644
--- a/src/tarballer.rs
+++ b/src/tarballer.rs
@@ -47,9 +47,11 @@ impl Tarballer {
         // Prepare the `.tar.gz` file.
         let gz = GzEncoder::new(create_new_file(tar_gz)?, flate2::Compression::best());
 
-        // Prepare the `.tar.xz` file.
+        // Prepare the `.tar.xz` file. Note that preset 6 takes about 173MB of memory
+        // per thread, so we limit the number of threads to not blow out 32-bit hosts.
+        // (We could be more precise with `MtStreamBuilder::memusage()` if desired.)
         let stream = xz2::stream::MtStreamBuilder::new()
-            .threads(num_cpus::get() as u32)
+            .threads(Ord::min(num_cpus::get(), 8) as u32)
             .preset(6)
             .encoder()?;
         let xz = XzEncoder::new_stream(create_new_file(tar_xz)?, stream);