exchange: improve computation of relevant markers for large repos
Find the candidates for nodes with relevant markers by going over
all markers instead of iterating over all nodes. Most nodes will
not have markers anyway.
Further optimize the code by allowing revsets as well, which reduces the
materialization cost.
# memorytop requires Python 3.4
#
# Usage: set PYTHONTRACEMALLOC=n in the environment of the hg invocation,
# where n>= is the number of frames to show in the backtrace. Put calls to
# memorytop in strategic places to show the current memory use by allocation
# site.
import gc
import tracemalloc
def memorytop(limit=10):
gc.collect()
snapshot = tracemalloc.take_snapshot()
snapshot = snapshot.filter_traces(
(
tracemalloc.Filter(False, "<frozen importlib._bootstrap>"),
tracemalloc.Filter(False, "<frozen importlib._bootstrap_external>"),
tracemalloc.Filter(False, "<unknown>"),
)
)
stats = snapshot.statistics('traceback')
total = sum(stat.size for stat in stats)
print("\nTotal allocated size: %.1f KiB\n" % (total / 1024))
print("Lines with the biggest net allocations")
for index, stat in enumerate(stats[:limit], 1):
print(
"#%d: %d objects using %.1f KiB"
% (index, stat.count, stat.size / 1024)
)
for line in stat.traceback.format(most_recent_first=True):
print(' ', line)
other = stats[limit:]
if other:
size = sum(stat.size for stat in other)
count = sum(stat.count for stat in other)
print(
"%s other: %d objects using %.1f KiB"
% (len(other), count, size / 1024)
)
print()