PR submission rate and review capacity

[carloea2]

What pace of PR submissions can the community comfortably support? Small, focused PRs help with individual reviews, but opening many at once also creates more requests for reviewers to track.

To make this discussion concrete, here are statistics for PRs opened in the 90 days ending September 23, 2026. There were 1,303 human-authored PRs from 40 contributors, excluding 186 automated PRs.

Submission rate

Contributor	PRs opened	Average/week	Opened in last 30 days	Still open
aglinxinyuan	364	28.3	53	27
mengw15	192	14.9	37	3
Yicong-Huang	128	10.0	1	2
kz930	95	7.4	53	27
carloea2	93	7.2	91	75
Ma77Ball	70	5.4	5	6
eugenegujing	59	4.6	3	3
yangzhang75	43	3.3	32	9
tanishqgandhi1908	43	3.3	20	0
PG1204	32	2.5	6	4

The overall rate decreased across the last two 30-day windows: 498 → 351 PRs. However, averages hide bursts. My own submissions include 58 PRs opened on August 31; 55 of my open, non-draft PRs had no submitted human review when checked.

PR sizes also differ. The median size of my PRs was 17 added/deleted lines, compared with approximately 228 for aglinxinyuan and 161 for mengw15. A PR count alone cannot describe the effort needed to review them.

Who handles the reviews?

Related to #8431, review work is concentrated among a few people.

Person	Distinct PRs reviewed	Different authors reviewed	PRs merged
aglinxinyuan	361	22	518
mengw15	329	19	187
Yicong-Huang	156	25	100
xuang7	155	20	92
carloea2	82	10	0
aicam	46	9	45
Ma77Ball	32	20	0
kunwp1	23	7	10

The top two account for 56% of reviewer–PR relationships and 73% of merges. The top four account for 82% of reviewer–PR relationships.

The author–reviewer relationships also show how much particular contributors depend on particular reviewers:

PR author	Most frequent reviewer	Second reviewer
aglinxinyuan	mengw15 — 256 PRs	Yicong-Huang — 58
mengw15	aglinxinyuan — 135	xuang7 — 19
Yicong-Huang	xuang7 — 50	aglinxinyuan — 44
kz930	carloea2 — 68	aglinxinyuan — 30
carloea2	Yicong-Huang — 16	aglinxinyuan — 7
Ma77Ball	aglinxinyuan — 49	Yicong-Huang — 6
eugenegujing	aglinxinyuan — 45	xuang7 — 6
yangzhang75	mengw15 — 24	aglinxinyuan — 3
tanishqgandhi1908	aicam — 33	mengw15 — 5
PG1204	xuang7 — 17	aglinxinyuan — 7

What is waiting?

Of the human-authored PRs opened during this period:

• 971 merged, 130 closed without merging, and 202 remain open.
• 37 of the open PRs are drafts.
• 89 open, non-draft PRs have no submitted human review; 79 of those are older than a week.

Among PRs that received a review, the median time to first human review was 8.6 hours. Among merged PRs, the median time to merge was 15.6 hours. These figures exclude unresolved waits, so they should be read alongside the queue numbers.

Questions for the community

1. Should contributors keep a manageable number of PRs awaiting review before opening more? Would a rough guideline help?
2. When several small changes are ready, is submitting them together useful, or should contributors coordinate the batch with reviewers first?
3. How can we distribute reviews more broadly when an author regularly depends on one or two people?
4. Would prioritizing existing PRs and reviewing others’ work help more than a weekly submission limit?

The numbers show different submission patterns and concentrated review work, but they do not establish what rate is sustainable. It would be useful to hear from the people doing the reviews about what is manageable.

Source and method: GitHub API metadata for apache/texera, June 25–September 23, 2026, with both boundaries at 23:02 UTC. Weekly averages include inactive weeks. Each human reviewer counts once per PR, excluding self-reviews and bots; ordinary conversation comments are not counted as reviews. Merge actions are counted separately and can include authors merging their own reviewed PRs. Related PRs and backports count separately. Open counts include only PRs created within this window, not the entire repository backlog.

[chenlica]

Here is some background about this discussion. Recently @carloea2 created a large number of PRs concurrently. People expressed at least two concerns:

• These PRs block other PRs in CI due to limited computing resources;
• Comments on some of the PRs can be applied to other PRs as well.

So it will be better to raise these PRs in small batches.

After seeing this discussion, a few members, including @carloea2 @aglinxinyuan @Xiao-zhen-Liu and me (@chenlica ) had an offline meeting and explained these concerns.

@carloea2 Please share your thoughts after the meeting. Thanks.

[carloea2]

Thanks. I agree with the concerns you summarized.

The offline discussion was helpful, and I understand that opening many PRs concurrently can put pressure on CI resources and make review feedback harder to apply consistently across related changes.

Going forward, I’ll submit changes in smaller batches and coordinate when I have several related PRs ready. I’ll also give existing PRs time to move through review before opening another large batch.

Thanks everyone for the feedback and for taking the time to discuss this.