wrangler r2 bucket list silently returns only the first 20 buckets, with no cursor, warning, or --per-page

[smashah]

Which Cloudflare product(s) does this pertain to?

Wrangler core, R2

What version(s) of the tool(s) are you using?

4.120.0 (also reproduced on 3.79.0)

What version of Node are you using?

22

What operating system and version are you using?

macOS 15 (Darwin 25.5.0, arm64)

Describe the Bug

wrangler r2 bucket list silently returns only the first 20 buckets, with no cursor, no "more results" indicator, no warning, and no flag to raise the page size. On an account with more than 20 buckets, the output is a truncated list that is indistinguishable from a complete one.

This is a quiet-wrong failure rather than a loud one, and it has a genuinely destructive natural consequence: if a bucket does not appear in the list, the obvious next step is to create it. On our account that would have meant creating a bucket over the top of one already holding 3.41 GB of live user-uploaded media.

The API is not the limitation. The same account, same credentials:

GET /client/v4/accounts/$ACCOUNT/r2/buckets?per_page=20   ->  20 buckets
GET /client/v4/accounts/$ACCOUNT/r2/buckets?per_page=100  ->  45 buckets

So the account genuinely has 45 buckets. Wrangler requests the default first page and presents it as the whole set.

Two smaller notes that make it harder to notice:

• The listing is alphabetical, so the omitted buckets are always the ones latest in the alphabet. Ours cut off at petrol-now-bucket-prod, and everything from q to z was invisible, including whisp-earth-production.
• The response body carries no result_info, so there is nothing in the payload a caller could inspect to detect truncation even if they wanted to.

Please provide a link to a minimal reproduction

No public repro link, since it requires an account with more than 20 R2 buckets. The reproduction is fully specified below.

Steps to reproduce

1. Use an account with more than 20 R2 buckets.
2. Run wrangler r2 bucket list.
3. Count the results: exactly 20, sorted alphabetically.
4. Query the API directly with per_page=100 and observe the true count.
5. Confirm a missing bucket really exists and is reachable with the same credentials:

$ wrangler r2 object get whisp-earth-production/definitely-not-a-real-key --file=/dev/null
✘ [ERROR] The specified key does not exist.

A key-level error proves the bucket resolved. The bucket was never missing, was never a permissions problem, and was reachable the entire time it was absent from bucket list.

Expected behaviour

Either of these would resolve it:

1. Paginate to completion by default, following the cursor until exhausted, so bucket list means what it says.
2. If that is undesirable for very large accounts, expose --per-page / --cursor and print an explicit truncation notice when more results exist, e.g. showing first 20 of N buckets.

Option 1 matches what users assume the command already does. Option 2 is acceptable as long as the truncation is visible, because the current failure mode is silent.

Additional context

Three separate people on our team reached the same wrong conclusion from this output on the same day, by three different routes, before someone checked the dashboard and found the bucket sitting there with 501 objects in it.

[workers-devprod]

Note

This is an automated, advisory triage report generated by workers-devprod. It is not an official maintainer response — a maintainer will follow up.

🤖 Automated triage report

Issue Triage: #15103

Summary

wrangler r2 bucket list silently returns only the first 20 buckets with no pagination, warning, or page-size flag, so a truncated list is indistinguishable from a complete one.

Classification

• Type: Bug
• Component: packages/wrangler (R2 code lives in wrangler)
• Severity: High
• Has Reproduction: Yes (no public repo link, but fully specified steps plus raw API evidence)
• Quality: Complete

Findings

• Created: not readable from context.json (single-line JSON truncated by tooling at ~2000 chars)
• Author: smashah
• Version: wrangler 4.120.0 (also reproduced on 3.79.0); Node 22; macOS 15 (arm64)
• Labels: not readable from context.json (see note below)
• Comments: not readable from context.json (see note below)

Analysis

• Component: The bug is in wrangler r2 bucket list. R2 command code lives in packages/wrangler, so this is a wrangler issue, not workerd or a platform issue.
• Root cause is plausible and well-evidenced: The reporter shows that the Cloudflare API itself returns 45 buckets with per_page=100 but 20 with per_page=20, and that wrangler only requests/prints the default first page. This points to wrangler not paginating (or not raising per_page) when listing R2 buckets.
• Severity is High, not cosmetic: This is a "quiet-wrong" failure. A missing bucket looks like it doesn't exist, and the natural next action (create the bucket) risks overwriting a live bucket holding user data. Silent, data-destructive potential elevates this above a normal display bug.
• Reproducibility: No public repro link (requires an account with >20 R2 buckets), but the report includes exact API calls, observed counts, and the alphabetical cut-off behavior. This is sufficient to act on; a maintainer can reproduce by seeding >20 buckets on a test account.
• Not closeable: Does not match any spam / duplicate / wrong-repo / stale / by-design / user-error category. It is a legitimate tooling bug with a clear fix direction (paginate the R2 bucket list, or at minimum raise per_page and surface truncation).

Tooling note: context.json is stored as a single line that exceeds the read/grep display limit (~2000 chars), so issue metadata that appears later in the JSON — labels, comment count/content, created/updated dates, and state — could not be confirmed. The recommendation below is based on the fully-readable body, which is complete enough to triage confidently. A maintainer should double-check current labels and any existing comments before acting.

Recommendation

Status: KEEP OPEN

Reasoning: This is a legitimate, high-quality, high-severity wrangler bug with clear API-level evidence and a plausible fix. The truncated listing can lead to accidental data loss, so it warrants a fix rather than closure.

Action: Confirm the R2 bucket list command does not paginate / defaults to a small per_page, then fix it to page through all buckets (or raise per_page and warn on truncation). Verify current labels/comments first, since those fields were not readable from the pre-fetched data.

Suggested Labels: package:wrangler, product:r2, bug

Suggested Comment

Thanks for the detailed report — the API evidence is really helpful. This does look like a bug in how wrangler r2 bucket list fetches buckets: it appears to request only the default first page and present it as the complete list, without paginating or surfacing that the results were truncated. We'll look into paginating through all buckets (and, at minimum, warning when there are more results). We'll follow up here with progress.

Structured summary (JSON)

{
  "issueNumber": 15103,
  "title": "wrangler r2 bucket list silently returns only the first 20 buckets",
  "githubUrl": "https://github.com/cloudflare/workers-sdk/issues/15103",
  "recommendation": "KEEP OPEN",
  "difficulty": "easy",
  "reasoning": "Legitimate high-severity wrangler bug: r2 bucket list does not paginate and silently truncates to 20 buckets, with clear API-level evidence and data-loss risk. Fix direction is clear.",
  "suggestedAction": "Fix wrangler r2 bucket list to paginate through all buckets (or raise per_page and warn on truncation); verify current labels/comments first.",
  "hasSuggestedComment": true,
  "suggestedLabels": ["package:wrangler", "product:r2", "bug"]
}

[dario-piotrowicz]

Thanks for the issue @smashah 😃

#15108 already has a fix for this! 💪