## The symptom
After upgrading Metabase from 0.49.x to 0.50.x, many deployments saw severe CPU
spikes and memory growth, with containers crashing every few hours from
out-of-memory. Downgrading to 0.49.12 made it go away. The thread (130+
comments) covered several symptoms: slow metadata endpoints, GC thrashing, and
OOM kills, mostly on larger dashboards.
## Why
This was a real regression in the 0.50 line, not a config problem. The Metabase
team traced the OOMs to a thread leak.
## The fix
Upgrade to Metabase 0.50.7 or newer. The thread leak fix shipped in PR #45253
in the 0.50.7+ releases; the maintainers asked everyone on the thread to
upgrade to 0.50.7 and then closed the issue once the fix was out.
## Workarounds people used in the meantime
- Downgrading back to 0.49.12 (made the symptoms go away).
- `MB_SQL_PARSING_ENABLED=false` (helped some setups but not the OOMs).
The durable fix is the patched release, not a config flag.