185
86%
master: 98%

Ran 23 Jul 2024 08:18PM UTC

Files 8

Run time 0s

Badge

Embed ▾

Committed 23 Jul 2024 08:13PM UTC coverage: 97.94%. Remained the same

Job # 185

Build Type

push

buildkite

Committed by

web-flow

Commit Message

always taking at least a read lock on shouldBubble (#33)

Previously, we were making a dirty read in `Get` when calling `shouldBubble` to see if we should bubble the item up. This creates a race condition with the write path. In cases where it was determined we should bubble up, we would take an exclusive lock, then check again. *Technically*, this is a race condition because we did make that dirty read, though we atoned for our sins with that second read.

Do we care? Well, maybe. The second read should make the action safe enough, but it does trip the race detector. There is no way to tell the race detector that this one is OK, so leaving this race condition in the code means that consumers of the library can't use the race detector for their own code. That's not an acceptable trade-off.

To avoid that, we can take a read lock before checking `shouldBubble`. If the answer is "no," we can just exit the function. If yes, we're going to have to take the exclusive lock anyway, so take the lock, check again, and bubble or not as appropriate. The impact of this lock is <0.5% when the cache is under 75% full and approximately 3% when the cache is over 3% full. That 3% represents only a few nanoseconds, so it's not as bad as it looks at first glance.

Coverage Stats

618 of 631 relevant lines covered (97.94%)

449.57 hits per line

TriggerMail / lazylru / 185 / 185
86%
master: 98%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job generic - 185

TriggerMail / lazylru / 185 / 185 86% master: 98%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job generic - 185

TriggerMail / lazylru / 185 / 185
86%
master: 98%

README BADGES
x