7 Jan 2021 |
underfenex | * Hi kt315 I've been doing some stress testing to a an Nginx Webserver. I had performance issues with Nginx so I updated 2 parameters in Nginx:
- keepalive_requests 100 (Which makes nginx process just 100 request and then close the connection)
- keepalive_timeout 60 (Keep the connection alive for a maximum of 60 seconds)
With these parameters I've been able to response with more QPS. However when I'm doing the stress testing I see that Closed (count and rate) start to show. In this context both Closes Count and rate are really errors? (because I make nginx force to close the connetions, or they are relevant errors?)
| 01:03:26 |
kt315 | You are right, they are not errors, probably should be yellow flags not red 🙂 | 01:06:08 |
underfenex | They are an indication that the webserver has closed the connection. I'm also seeing that after the normal test finishes It seems that Stressgrid keeps emiting request, could it be due to trying to recover these "failed connections"? | 01:15:01 |
kt315 | yes, there's usually some inertia when connections are not closed by the client side | 01:16:33 |
underfenex | Great! Thnx! | 01:17:10 |
underfenex | Awesome tool! | 01:17:26 |
kt315 | you are very welcome and thank you for kind words | 01:17:47 |
underfenex | Regarding this Client side zombie connections inertia, will that happend with real users (Chrome or another browser will create this inertia)? Or it's just regarding the tool? | 01:20:33 |
kt315 | only with the tool, Chrome will not reconnect unless there is a new request | 02:11:47 |
8 Jan 2021 |
underfenex | Hi again! I've stresstesting my phoenix app, and I get around 500 req/s with SSL termination and CPU Usage of around 100% in an Azure B1s (1 CPU and 1GB RAM), the weird issue is that:
- When I made the load just HTTP (Not SSL termination) I though it would go faster, but the same thing happen. It's weird because because it should be faster as no SSL termination is done.
- It can process more request per second, but the latency starts ramping.
- I used these recomendations https://www.theerlangelist.com/article/phoenix_latency (except the part that builds the relesea, I'm doing the test just with mix phx.server)
| 14:36:31 |
9 Jan 2021 |
kt315 | generally TLS/SSL is very fast unless you are sending very large payloads, can you look what creates that 100% CPU usage? | 03:01:13 |
31 Jan 2021 |
| bxo set a profile picture. | 05:31:15 |
6 Feb 2021 |
underfenex | Hi kt315_ the only consumption is from beam, is there a way to check exactly in detail this consumption? | 14:51:25 |
9 Feb 2021 |
kt315 | https://www.youtube.com/watch?v=xi-ixFPjNpc | 02:37:16 |
kt315 | ^^ this is good talk--hopefully, it will give you some pointers | 02:37:51 |
5 Jul 2021 |
| Andres Hernandez joined the room. | 20:03:51 |
6 Oct 2021 |
| Slack Integration | 08:34:11 |
18 Apr 2022 |
| @wybpip:matrix.org joined the room. | 12:37:59 |
| @wybpip:matrix.org left the room. | 12:38:00 |
2 May 2022 |
| @wybpip:matrix.org joined the room. | 00:43:15 |
| @wybpip:matrix.org left the room. | 00:43:16 |
12 Jun 2022 |
| nemlazukku joined the room. | 04:48:41 |
25 Jun 2022 |
| @better_sleeping:converser.eu joined the room. | 13:09:17 |
| @better_sleeping:converser.eu left the room. | 13:09:26 |
29 Apr 2023 |
| @suwako:hot-chilli.im joined the room. | 12:19:15 |
30 Apr 2023 |
| imlostlmao joined the room. | 18:30:22 |
11 May 2023 |
| @suwako:hot-chilli.im removed their profile picture. | 13:01:46 |
| @suwako:hot-chilli.im left the room. | 13:41:39 |
5 Jan 2024 |
| imperialstar joined the room. | 22:53:41 |
imperialstar | Hey kt315_ , I want to use stressgrid for stresstesting GraphQL subscriptions. My questions are
-
Do you maintain the project? I see that the last commit is OTP23 support. If I add OTP26 support, will you be able to review it?
-
Are there any considerations around websocket support?
| 22:55:29 |