Fix misleading decompression errors for missing URLs in glob patterns

[alexey-milovidov]

Summary

• When a glob URL pattern expands to URLs that return 404, http_skip_not_found_url_for_globs (enabled by default) silently swallows the HTTP error. But the decompression wrappers (Zlib, Brotli, LZMA) didn't handle a completely empty input stream — they attempted decompression with zero bytes and produced misleading errors like "inflate failed: buffer error" instead of simply returning no rows.
• Fixed ZlibInflatingReadBuffer, BrotliReadBuffer, and LZMAInflatingReadBuffer to detect empty streams (total_in == 0 and inner buffer at EOF) and return EOF. Zstd and Lz4 already handled this correctly.
• Added a test covering all 5 compression formats (gzip, zstd, brotli, lz4, lzma).

#49231 (comment)

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Fix misleading "inflate failed: buffer error" when reading non-existent compressed files via url() table function with glob patterns. Now returns empty result as expected when http_skip_not_found_url_for_globs is enabled.

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches core decompression read paths used by url() and other inputs; a small logic change could affect EOF/error handling for truncated or unusual streams across multiple codecs.

Overview
Fixes misleading decompression failures when a globbed url() input is skipped due to 404s (via http_skip_not_found_url_for_globs) by treating a completely empty underlying stream as EOF instead of attempting to decompress.

Adds empty-stream detection to ZlibInflatingReadBuffer, LZMAInflatingReadBuffer, and BrotliReadBuffer (tracking total_in for brotli) and introduces a stateless test that queries missing .gz/.zst/.br/.lz4/.xz URLs and expects zero rows without errors.

^{Written by Cursor Bugbot for commit 350d87b. This will update automatically on new commits. Configure here.}

[clickhouse-gh]

Workflow [PR], commit [5ff912f]

Summary: ✅

---

AI Review

Summary

This PR fixes misleading decompression errors for empty streams returned by missing URLs in glob-based url reads by adding explicit empty-input handling in Brotli, LZMA, and Zlib inflating buffers, and adds a stateless regression test that verifies zero-row behavior across compressed URL suffixes. I did not find any high-confidence correctness, safety, or performance regressions in the touched code.

Missing context

• ⚠️ No CI run artifacts/logs were provided in this review context.

ClickHouse Rules

Item	Status	Notes
Deletion logging	➖
Serialization versioning	➖
Core-area scrutiny	✅
No test removal	✅
Experimental gate	➖
No magic constants	✅
Backward compatibility	✅
SettingsHistory.cpp	➖
Safe rollout	✅
Compilation time	✅

Final Verdict

• Status: ✅ Approve

[alexey-milovidov]

But why we are even trying to decompress on 404?

[alexey-milovidov]

But why we are even trying to decompress on 404?

We are not decompressing the 404 response body. The flow is:

1. ReadWriteBufferFromHTTP is created with delayed initialization (delay_initialization = true because current_uri_options.size() == 1 in StorageURLSource).
2. FormatFactory::getInput wraps it with wrapReadBufferWithCompressionMethod (gzip decompression based on the .gz URL extension). At this point, no HTTP request has been made yet.
3. When the format reader tries to read, the chain is: Format → GzipDecompression → ReadWriteBufferFromHTTP.
4. The decompression buffer calls in->nextIfAtEnd(), which triggers ReadWriteBufferFromHTTP::nextImpl() → initialize() → HTTP request → 404.
5. Because http_skip_not_found_url_for_globs is true, the 404 is caught at ReadWriteBufferFromHTTP.cpp:482 and the buffer returns EOF (empty).
6. The decompression buffer now has zero bytes of input and tries to run inflate() / lzma_code() / BrotliDecoderDecompressStream() with empty input → fails with a misleading error.

So the decompressor sees an empty stream, not the 404 error page. The fix handles this edge case in the decompression buffers. The alternative would be to propagate the "not found" status through the buffer chain so the decompression wrapper can be skipped entirely, but that would be more invasive — it would require changes to CompressedReadBufferWrapper, FormatFactory, and StorageURLSource.

[alexey-milovidov]

LGTM

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	83.70%	83.80%	+0.10%
Functions	23.90%	23.90%	+0.00%
Branches	76.20%	76.30%	+0.10%

PR changed lines: PR changed-lines coverage: 91.67% (33/36, 0 noise lines excluded)
Diff coverage report
Uncovered code

[zlareb1]

@alexey-milovidov the fix should be made for bz2 as well: #99806