poireau

Recently saw this through https://discuss.systems/@pkhuong/114532017716618023 which summaries why this would be useful: > Flagging old live (i.e., potentially immortal) allocations is easier (https://github.com/backtrace-labs/poireau) and more closely matches the goal of identifying why the heap grows instead of entering a steady state. Ej. Why is Sidekiq memory consumption growing over time.

Among the less niche ones: - UMASH: https://github.com/backtrace-labs/umash Fast hash functions don't usually come with collision bounds, and crypto hashes try too hard for my needs. With umash, we get speed that's comparable to regular hash functions, and a 128 bit output such that hardware failure is more likely than collisions. - poireau: https://github.com/backtrace-labs/poireau existing heap profilers are more intrusive (first, you have to replace the allocator), and, in long-lived servers, any long-lived allocation is cause for concern, regardless of reachability. A flamegraph compatible output is also useful for visualisation. - CSM: https://github.com/pkhuong/csm Testing an SLO-type property that's expected to hold, e.g., 99% of the time, quickly becomes an annoying exercise in balancing test duration and false alarms. CSM impleme