core/src/main/java/io/questdb/cutlass/http/processors/LineHttpProcessorState.java (1)

185-217: Fix gzip inflate handling when the buffer fills up.

Zip.inflate can return Z_BUF_ERROR after writing bytes once the receive buffer runs out of space. Because bufPos is only advanced after the negative-return check, this branch always falls through to MESSAGE_TOO_LARGE, dropping valid gzip payloads while the parser never sees the bytes that were already produced. Move the bufPos update ahead of the status check (compute produced = len - Zip.getAvailOut(...)), process that chunk, and only reject after trying to compact/grow the buffer—mirroring the uncompressed path. This is the same issue I raised earlier; it still needs to be fixed.

-            int ret = Zip.inflate(inflateStream, p, len, false);
-
-            if (ret < 0) {
-                if (ret != Zip.Z_BUF_ERROR) {
+            int status = Zip.inflate(inflateStream, p, len, false);
+            int produced = len - Zip.getAvailOut(inflateStream);
+            if (produced > 0) {
+                recvBuffer.setBufPos(p + produced);
+            }
+
+            if (status < 0) {
+                if (status != Zip.Z_BUF_ERROR) {
                     reject(ENCODING_NOT_SUPPORTED, "gzip decompression error", fd);
                     cleanupGzip();
                     return;
                 }
 
-                if (recvBuffer.getBufPos() > pp) {
-                    // when inflate fails with Z_BUF_ERROR - this means either
-                    // have not processed buffer yet, or it is too small. Trigger buffer processing
-                    // to make space in the buffer
-                    currentStatus = processLocalBuffer();
-                    continue;
-                }
-
-                reject(MESSAGE_TOO_LARGE, "server buffer is too small", fd);
-                cleanupGzip();
-                return;
+                if (produced > 0) {
+                    currentStatus = processLocalBuffer();
+                    pp = recvBuffer.getBufPos();
+                    continue;
+                }
+                if (!recvBuffer.tryCompactOrGrowBuffer()) {
+                    reject(MESSAGE_TOO_LARGE, "server buffer is too small", fd);
+                    cleanupGzip();
+                    return;
+                }
+                continue;
             }
 
-            recvBuffer.setBufPos(p + (len - Zip.getAvailOut(inflateStream)));
-
-            if (ret == Zip.Z_STREAM_END) {
-                currentStatus = processLocalBuffer();
-                cleanupGzip();
-                break;
-            }
+            if (produced > 0) {
+                currentStatus = processLocalBuffer();
+                if (stopParse()) {
+                    return;
+                }
+                pp = recvBuffer.getBufPos();
+            }
+
+            if (status == Zip.Z_STREAM_END) {
+                cleanupGzip();
+                break;
+            }

core/src/main/c/share/zip.c (2)

136-140: Duplicate API: getAvailOut duplicates availOut.

Zip_availOut(ptr) already exposes avail_out. Consider reusing it (or aliasing on Java) to keep a single native symbol and avoid drift.

Option A: remove getAvailOut and use availOut on Java.
Option B: keep both but document one as preferred to avoid confusion.

---

90-107: Optional: minor JNI hygiene.

• Return -1 on OOM is fine but collides with Z_ERRNO; consider a distinct sentinel like INT_MIN for clearer differentiation on Java side.
• Consider factoring stream allocation into a tiny static helper to reduce repetition in inflateInit/inflateInitGzip.

Also applies to: 128-134

compat/src/test/java/io/questdb/compat/InfluxDBClientTest.java (2)

225-235: Add negative-path gzip tests.

Consider adding:

• Corrupted gzip body → 400 with parse error.
• Mismatched header (Content-Encoding: gzip, uncompressed body) → 400/415.
• “Gzip bomb” expansion exceeding line.http.max.recv.buffer.size → request too large.

I can draft these tests with minimal scaffolding if helpful.

---

776-785: Avoid potential barrier hangs: add timeout.

Use a bounded await to fail fast on thread errors and avoid indefinite waits.

-                        startBarrier.await();
+                        startBarrier.await(60, java.util.concurrent.TimeUnit.SECONDS);

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

core/src/main/java/io/questdb/cutlass/http/processors/LineHttpProcessorState.java (2)

178-228: Decompress more before parsing to reduce parser fragmentation.

Inflate until either stream end or the local buffer can’t accept more bytes, then call processLocalBuffer() once. This reduces parse calls and improves throughput.

---

178-228: Handle Z_BUF_ERROR by compacting/growing buffer before rejecting.

Currently, if the output buffer is full and inflate returns Z_BUF_ERROR with newBytes == 0, the code immediately returns 413. This breaks valid requests whose decompressed size fits within line.http.max.recv.buffer.size. Mirror the uncompressed path: try recvBuffer.tryCompactOrGrowBuffer() and retry before failing.

-            if (ret < 0) {
-                if (ret != Zip.Z_BUF_ERROR) {
-                    reject(ENCODING_NOT_SUPPORTED, "gzip decompression error", fd);
-                    cleanupGzip();
-                    return;
-                }
-
-                // inflate can return Z_BUF_ERROR after writing bytes once the recv buffer runs out of space
-                if (newBytes > 0) {
-                    currentStatus = processLocalBuffer();
-                    pp = recvBuffer.getBufPos();
-                    continue;
-                }
-
-                reject(MESSAGE_TOO_LARGE, "server buffer is too small", fd);
-                cleanupGzip();
-                return;
-            }
+            if (ret == Zip.Z_BUF_ERROR) {
+                // Output buffer exhausted.
+                if (newBytes > 0) {
+                    currentStatus = processLocalBuffer();
+                    pp = recvBuffer.getBufPos();
+                    if (stopParse()) {
+                        cleanupGzip();
+                        return;
+                    }
+                    continue;
+                }
+                if (recvBuffer.tryCompactOrGrowBuffer()) {
+                    // Retry inflate with more space.
+                    continue;
+                }
+                reject(MESSAGE_TOO_LARGE, "transaction is too large, either flush more frequently or increase buffer size \"line.http.max.recv.buffer.size\"", fd);
+                cleanupGzip();
+                return;
+            } else if (ret < 0) {
+                // Other zlib errors
+                reject(ENCODING_NOT_SUPPORTED, "gzip decompression error", fd);
+                cleanupGzip();
+                return;
+            }

compat/src/test/java/io/questdb/compat/InfluxDBClientTest.java (2)

733-734: Use locale‑independent uppercasing to avoid surprises.

Case transforms can differ by default locale (e.g., Turkish). Prefer ROOT.

Apply:

-        String tableNameUpper = tableName.toUpperCase();
+        String tableNameUpper = tableName.toUpperCase(java.util.Locale.ROOT);

If you prefer an import:

import java.util.Locale;
// ...
String tableNameUpper = tableName.toUpperCase(Locale.ROOT);

---

787-836: Parallel helper is solid; barrier start avoids head‑of‑line bias.

Minimal suggestion: name threads for easier log triage.

Example:

-                threadList.add(new Thread(() -> {
+                threadList.add(new Thread(() -> {
                     // ...
-                }));
+                }, "ilp-http-" + threadNo));

core/src/test/java/io/questdb/test/cutlass/http/line/LineHttpFailureTest.java (2)

492-496: Close GZIPOutputStream with try‑with‑resources to guarantee trailer flush.

Use TWR instead of manual finish(); it’s safer and clearer.

-ByteArrayOutputStream out = new ByteArrayOutputStream();
-GZIPOutputStream strm = new GZIPOutputStream(out);
-strm.write(b);
-strm.finish();
-byte[] outBytes = out.toByteArray();
+ByteArrayOutputStream out = new ByteArrayOutputStream();
+byte[] outBytes;
+try (GZIPOutputStream strm = new GZIPOutputStream(out)) {
+    strm.write(b);
+}
+outBytes = out.toByteArray();

---

473-522: Add coverage for Content‑Encoding edge cases.

Please add tests for:

• Case‑insensitive header (e.g., “GZip”).
• Multiple codings (e.g., “gzip, deflate” and “deflate, gzip”) → expect 415 until multi‑step decoding is supported.

core/src/main/java/io/questdb/cutlass/http/processors/LineHttpProcessorState.java (1)

117-122: Optionally reset the gzip flag on cleanup to avoid accidental reuse.

Not required, but harmless and clarifies lifecycle.

     public void cleanupGzip() {
         if (inflateStream != 0) {
             Zip.inflateEnd(inflateStream);
             inflateStream = 0;
+            isGzipEncoded = false;
         }
     }

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro