DurableHttpUsingFileSizeRolledBuffers - interval flush period doubts #164
Replies: 1 comment 4 replies
-
Hi @sheltertake and welcome to the repo! I would like to propose that we break down you questions. I'll try to answer them one at a time, and when we both feel the question is answered we'll proceed with the next question in your post, all the way down to your final question. Would that be alright with you?
Let's start with this question, shall we? What do you mean when you say that "the events arrive broken"? Let me start by building the foundation for our discussion. When you write a log event using Serilog, what will happen is that Serilog will pass that event to this sink. This sink will then proceed to serialize the CLR log event to a textual representation using a
Looking at your sample code it seems like you are using the Let's summarize this, my two initial questions are:
|
Beta Was this translation helpful? Give feedback.
-
Hi @FantasticFiasco ,
Last friday I start playing with this sink. I want to replace the way I log now applications (terrible). So I'm studying a bit.
From what I understood... but everything I write here can be wrong by my limited experience on serilog and elk.
console application very similar of your sample
api setup
api recipe
repo
Benchmark results are not the problem. The wrk setup could be too much aggressive and I didn't setup a warmup phase and most of all, the benchmark is not honest because the serilog-elk scenario have an elk cluster up and running on my poor laptop in concurrency with same resources. The other ones not.
The problem is the scenario produce in 20s from 70k to 100k events. And these events arrive to elk very very slowly.
2 minutes between each packet. I tried playing with "period:" option but nothing change. I tried to raise batchPostingLimit: to 10000. I collected in this way 40k events in 6-8 minutes after benchmark finish.
It's a behavior explainable? Is a sort of exponential retry alghoritm of serilog that feels api very stressed and schedule logs flush with interval very lazy? switching off api cause flush but it seems "one" packet more sent. Not all it has on the file.
Thank you for any kind of attention you want to give to this post.
I tried to write it down as best as I can.
Max
Beta Was this translation helpful? Give feedback.
All reactions