Travis Downs

> Regarding compiler instrumentation: I really don't know about the difficulty of modifying gcc/LLVM for this purpose, maybe it's actually hard. I think in theory such instrumentation should be fairly...

> I'm not sure allocating a register gets you anything...? `add reg, imm` also modifies flags, and each basic block gets an independent counter anyways. Without preserved flags, it should...

@michael-redpanda the kind of loop I was talking about that's efficient is something like this: ``` template ss::future nice_for_each(Iterator begin, Iterator end, Fn f) { while (begin != end) {...

Can `throw_if_modified(version_snapshot);` be checked once after the all the looping is complete? This would expand the options to use a helper here.

Sorry I forgot the godbolt link: https://godbolt.org/z/qGecYf5e9

@mmaslankaprv over here I put a more fleshed out version of the helper: https://github.com/redpanda-data/redpanda/pull/16662 The idea is you could replace your inner loop with async_for_each_counter, like so: ``` async_counter work_counter;...

Some numbers from local testing of `{std::vector, frag_vector}` x `{std::sort, std::nth_element}` combinations (look at `median` column): ``` single run iterations: 0 single run duration: 1.000s number of runs: 1 number...

@mmaslankaprv I learned overnight that simply creating/destroying the 23k partitions is not enough, this was working fine when run 100s of times in a loop. So the actual thing I...

> Interesting thing is that these partitions are seen in the logs (grep for test-topic-) and each shard hosts 300-400 partitions of a single topic: It did strike me that...

@mmaslankaprv - we ended discussing this a bit on slack and I think the current status is that you have the timed watchdog checked in to give more info when...