Lessons in Managing Haskell Memory

At Channable, we process several billion product data records for our customers every day according to user-customizable rules. Internally, this work is subdivided into jobs, where one job takes a set of products and the rules for how to process them as input, and returns the transformed set of products as output. Those datasets range from a few dozen to tens of millions of products in size. The a

[igrep]

どうやってGCにかかる時間を減らしたか。パースしてできる文字列が共通していることが多いのでcompact regionに入れたり、compact regionを入れるハッシュテーブルを強引にunboxedにしたり、mutableな配列を減らしたり。