9
9
10
10
This is a good time to do it because This Week in Rust (see below) is usually
11
11
published on Tuesday, US time, and so it means the PR to include the triage
12
- details in TWiR can be merged shortly before publication.
12
+ details in TWiR can be merged shortly before publication. This time is also
13
+ shortly before the weekly Rust compiler meeting, where the results are looked
14
+ at.
13
15
14
16
## Instructions
15
17
16
- Check the most recent triage log entry. Look for responses in PRs, and follow
17
- up on any promised actions. (I.e. nag people!)
18
+ First, check the previous triage log entry. Look for responses in PRs, and
19
+ follow up on any promised actions. (I.e. nag people!)
18
20
19
- Start a new triage log entry. Follow the format of the preceding entries.
21
+ Start the new triage log entry in a new file using a ` YYYY-MM-DD.md ` -form name.
22
+ Follow the format of the previous entries.
20
23
21
- Look through the [ ` instructions:u ` graphs] ( https://perf.rust-lang.org ) for each
22
- benchmark.
23
- - Record the hash of the latest measured revision in the log entry.
24
- - Set the "start" revision to the previous log entry's latest revision, and the
25
- "end" revision to the latest revision. Full SHAs must be used.
24
+ View the [ perf website] ( https://perf.rust-lang.org ) .
25
+
26
+ - Determine the revision range. The start revision corresponds to the end
27
+ revision from the previous log entry. The end revision is the latest measured
28
+ revision on perf.rust-lang.org.
29
+ - Set the range on the graphs by setting the "start" and "end" revisions at
30
+ the top of the page. Full SHAs must be used.
26
31
- Uncheck the "Absolute data" checkbox, because that makes changes easier to
27
32
see.
33
+ - Record the revision range, with a link, in the log entry.
28
34
29
- Look for significant changes (regressions or improvements) in the graphs.
35
+ Look for significant changes (regressions or improvements) in the instruction
36
+ count graphs.
30
37
- Click and drag a region of a graph to zoom in on it. This is useful when data
31
38
points are close together.
32
39
- Click on a data point to open the "compare" page for that merge.
33
40
- Click on the "compare" link at the top of the measurements on that page to
34
41
open the page of commits in the merge.
42
+ - Alternatively, it may be easier to simply click through to the "compare" page
43
+ for every revision in the range (there often aren't that many), rather than
44
+ looking at the graphs.
35
45
- A change isn't significant unless one or more of the benchmarks changed by at
36
46
least 1%.
37
47
@@ -42,10 +52,11 @@ Easy cases: there is only a single PR in the merge.
42
52
regression, consider requesting a backout. It may be worth looking through
43
53
the comments to see if any perf CI runs were done, and whether the
44
54
regression was expected.
45
- - Add an entry to the triage log. Include the PR title, a link to it, and a
46
- link to the performance results. Include useful details, such as the size of
47
- the regression/improvement, and any promises of follow-up action from authors
48
- in the case of a regression.
55
+ - Add an entry to the triage log. Include the PR title and number, a link to
56
+ the PR comment you added mentioning the performange effect, and a link to the
57
+ performance results. Include useful details, such as the size of the
58
+ regression/improvement, and any promises of follow-up action from authors in
59
+ the case of a regression.
49
60
50
61
Difficult cases: the merge was a rollup of multiple PRs.
51
62
- Look through the PRs and try to determine which was the cause. Often you
@@ -72,10 +83,6 @@ Rust](https://github.com/emberian/this-week-in-rust/).
72
83
- That item should be a link to the triage log entry with the form
73
84
"YYYY-MM-DD", possibly with some brief text about notable things.
74
85
75
- If you have any questions, the #wg-compiler-performance channel on Discord and
76
- the t-compiler stream on Zulip are good places to ask.
77
-
78
- ## Triage logs
79
-
80
- - [ 2020] ( 2020.md )
86
+ If you have any questions, the ` t-compiler/performance ` stream on Zulip is the
87
+ best place to ask.
81
88
0 commit comments