HAProxy replies client with RST when client side reuses port in http mode and soft reload is issued on HAProxy
還沒有人認領這個 Issue。
評估
- 難度
- 4/5
- 預估耗時
- 3-5 天
- 新手友好度
- 48/100
- Issue 類型
- 缺陷
- 描述清晰度
- 基本清楚
- 活躍度
- 冷清
- 技術堆疊
- c
- 領域
- backend, networking
研究方向
Start by reproducing the issue with HAProxy master-worker mode, the shown HTTP backend settings, connection pooling, and the specified keep-alive timeouts. Use the described tcpdump observations to trace the client connection across a soft reload; done means reused client requests continue without HAProxy sending FIN or RST.
由索引模型根據 Issue 內容生成。
描述
Detailed Description of the Problem
Setup
Clients --> Gateway -> HAProxy -> Backend Server
Client side Gateway is using ConnectionPools which means the client tries to reuse exiting connection (5 sec timeout).
Client side keep alive timer: 5 seconds
HAProxy keep alive timeout: 60 seconds
HAProxy Version haproxy-3.0.17
Issue
Whenever client reuses existing connection towards HAProxy, and HAProxy is issued soft reload (due to changes to in the ingress or backend), then the HAProxy terminating the client connection with RST.
I believe above is not an expected behaviour on the HAProxy. Either the soft reload or client side connection reuse should not lead to termination of the existing connection by the HAProxy.
Current Analysis
Analysed the tcpdumps and client connection behaviour for the above scenario.
Tcpdump traces indicate that whenever clients have started the connections:
- There is single tcp/SSL connection from client to the HAProxy for entire duration
- There are multiple http connections from HAProxy to the backend server - one for each client request
When reload is issued on HAProxy the following observation is made:
- No issue or tcp/http packets seen between HAProxy and the backend server
- If client tries to send another request using the existing tcp connection, HAProxy is replying with FIN and RST
Expected Behavior
The client requests should continue to work as it was working prior to the reload of the HAProxy.
Steps to Reproduce the Behavior
- Connect some Gateway before the HAProxy with connection pooling enabled, set the port reuse timeout to 5 seconds
- Use standard master-worker mode on HAProxy, add a HTTP backend, set http-keep-alive time out to 1m
- Establish multiple http GET requests from the client side and while the requests are in progress, issue soft reload on the HAProxy
Do you have any idea what may have caused this?
No response
Do you have an idea how to solve the issue?
No response
What is your configuration?
**HAProxy Config**
global
master-worker
defaults
log global
option redispatch
option http-server-close
option http-keep-alive
timeout client 600s
timeout client-fin 600s
timeout connect 5s
timeout http-keep-alive 1m
timeout http-request 5s
...
backend apps
mode http
balance hdr(xyz)
option srvtcpka
...
Output of haproxy -vv
NA
Last Outputs and Backtraces
Additional Information
No response
- 主要語言
- C
- 星號
- 6.9k
- 分支
- 976
- PR 合併指標
- 30 天內沒有已合併 PR
環境準備
這個專案沒有提供開發容器、Dockerfile 或貢獻指南，環境需要你自己搭建：先看它的 README，通用步驟見我們的新手貢獻指南。
從這裡開始
- 先讀完整個 Issue，再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫，在一個分支上完成修改。
- 送出 Pull Request，並在描述裡引用這個 Issue 編號。
haproxy/haproxy 的其他 Issue
-
type: feature
難度 1/5 1 小時以內 新手友好度 77/100
-
難度 2/5 1-3 小時 新手友好度 82/100
-
難度 2/5 1-3 小時 新手友好度 82/100
-
type: code-report
難度 2/5 1-3 小時 新手友好度 74/100
-
type: code-report
難度 1/5 1 小時以內 新手友好度 82/100
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 62/100
-
難度 2/5 1-3 小時 新手友好度 72/100
fluent/fluent-bit#12579 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 72/100
obsproject/obs-studio#14036 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 68/100
維護者通常 1 天內回覆
-
難度 1/5 1 小時以內 新手友好度 72/100