- 06 9月, 2019 1 次提交
-
-
由 Nick Little 提交于
-
- 04 9月, 2019 2 次提交
-
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
- 31 8月, 2019 1 次提交
-
-
由 Magnus Edenhill 提交于
When control messages were skipped the remaining buffer was not consumed which lead to the messageset parser reading the remaining bytes as messages, which in best case resulted in parse errors.
-
- 30 8月, 2019 1 次提交
-
-
由 Matt Howlett 提交于
-
- 28 8月, 2019 2 次提交
-
-
由 Magnus Edenhill 提交于
-
由 Matt Howlett 提交于
-
- 20 8月, 2019 6 次提交
-
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
(cherry picked from commit 06b0b149)
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
Directly access the slice's segments, rather than using the more costly abstraction. This also fixes a bug where varint decoding would truncate/wrap numbers above SIZE_MAX, which is just 32-bits on 32-bit platforms.
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
- 19 8月, 2019 5 次提交
-
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
- 16 8月, 2019 4 次提交
-
-
由 Magnus Edenhill 提交于
This makes sure that a consumer_poll() call after pause() will not return any messages.
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
-
由 Matt Howlett 提交于
-
- 15 8月, 2019 1 次提交
-
-
由 Matt Howlett 提交于
-
- 14 8月, 2019 7 次提交
-
-
由 Chris Broadbent 提交于
* Fix UB call to memcpy rd_tmpabuf_write is sometimes called with a NULL buffer to clone, leading to UB when calling memcpy. It doesn't look like the func could just return NULL in this case either. Almost all the callers ignore the possibility of NULL, or are interpreting NULL as an allocation failure. * Fix call to qsort that triggers UBSan * Reworked UBSan fixes based on review feedback
-
由 Vladimir Sakharuk 提交于
Change-Id: Ib0aab2e663a6ec63426605f4bfb527a6ee95fd94
-
由 Nikhil Benesch 提交于
Like Linux, macOS supports the nonstandard pthread_setname_np, though the signature is slightly different. Learn to detect and use macOS variants to, to make multithreaded debugging on macOS a bit easier.
-
由 Magnus Edenhill 提交于
https://github.com/confluentinc/confluent-kafka-dotnet/issues/1027 This also improves the granularity of message timeout scans to match the actual message timeout. (#2202) The minimum effective message timeout will still be 1000ms, but the granularity after the initial 1000ms is now on millisecond precision.
-
由 Magnus Edenhill 提交于
Consumer transaction support
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
- 13 8月, 2019 10 次提交
-
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Matt Howlett 提交于
-
由 Magnus Edenhill 提交于
-
由 Magnus Edenhill 提交于
This saves quite a few memory barriers for high-performance Producers.
-