cargo profiler has been running for 35 minutes for a test that should take ~3 minutes. when do I give up?
-
Show this thread
-
stopping it now that it's been running for an hour
1 reply 0 retweets 1 likeShow this thread -
it finished in release mode, but the results are very unhelpfulpic.twitter.com/beXCLhbysd
1 reply 0 retweets 1 likeShow this thread -
first is 16k records, 2nd is 32k records. in each case, the average size of a data block is ~2. pretty bad, that should be a few hundred at leastpic.twitter.com/vzFl9RIrjk
1 reply 0 retweets 1 likeShow this thread -
the write cache speeds things up about 3x already which is good, but I'll need about a 10x improvement in these write speeds to be able to process planet.osm in <24h on a single cheap vps instance
2 replies 0 retweets 0 likesShow this thread -
Replying to @substack
Maybe https://github.com/ferrous-systems/flamegraph … might be another useful tool to try out; get some CPU profiles going to see where some bottlenecks might lie.
1 reply 0 retweets 1 like -
Replying to @yoshuawuyts @substack
Renée #ZwartePietIsRacisme Retweeted Renée #ZwartePietIsRacisme
oh i should use this instead of doing https://twitter.com/goto_bus_stop/status/1111390834086154240 … lol
Renée #ZwartePietIsRacisme added,
Renée #ZwartePietIsRacisme @goto_bus_stopReplying to @substack`perf` works okay for rust stuff in my experience cargo test --release perf record --call-graph dwarf cargo test --release perf script | FlameGraph/stackcollapse-perf.pl --all | FlameGraph/flamegraph.pl --color=hot --hash > flame.svg gives a nice ish svg (lots of test noise)1 reply 0 retweets 1 like -
Replying to @goto_bus_stop @substack
haha, for sure. Tho additionally: *always* remember to set a sample rate. E.g. $ perf record -F 997 -g "cargo bench" Setting a non-even number also prevents from accidentally only profiling the wrong data at a nice interval. This does almost 1000 samples / sec.
1 reply 0 retweets 1 like -
Replying to @yoshuawuyts @substack
is a round sample rate also potentially problematic if the code is not using timers or anything, but only doing sequential number/io crunching?
1 reply 0 retweets 1 like
yep! -- even if your code might not be on a nice round number, the underlying runtime (node) or OS might be. Using a somewhat erratic sample rate makes for better profiles.
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.