Our team has been using github action for CI/CD for a while. It’s very convenient and easy to use. We use it to build docker images, push to docker registry and deploy to kubernetes cluster. But there is one thing that we are trying to improve, that is we want to have a manual approval step before deploying to production. We tried many methods but none of them works well until we found this manual-workflow-approval action.
As I mentioned in my previous post I have setup a kubernetes(k3s) cluster on my home server. I was thinking how I could use this then I decided to host a blog site using the most used CMS on the internet, Wordpress. Here I would explain how I did it.
I have just built a sql chatbot for local data using langchain and streamlit. I found the whole process surprisingly smooth with these two great tools, I put here how I built this and some thoughts alongside.
In case you don’t know what langchain and streamlit are, here is a brief introduction. Langchain in my mind is a tool to connect to all kinds of language models(e.g. openai) and provide interfaces to easily using them. Streamlit is a tool that allows you to write UI included apps with only python. so langchain+streamlit means you can build UI included llm apps with only python.
I was working on testing upload images to Cloudflare using their API. The task was supposed to be simple because cloudfare have some code samples here, however I was having trouble getting it to work. As I’m using node+axios , the sample code on document is like
I will explain in this article how to deploy and serve a kubenetes cluster(k3s) using your home laptop/PC. I will also show you how to expose your services to the internet with cloudflare tunnel. I will use a simple web app as an example to show how to build a full-stack application and serve it with the home server.
Here is the overview of the architecture setup.
The laptop running all these stuff is a 7 year old Thinkpad with Interl i5 + 16G Mem installed with Ubuntu
[This is a note first published internally, I copied here for backup ]
Although I have deployed a couple of apps in our k8s cluster, I find it still tricky to go through the whole journey fluently. I took some notes and put them down here for a memo. Hope it helps developers like me who are still experiencing k8s issues from time to time.
The deployment steps
So you have a Dockerfile in your project and a docker engine running on your machine
Kong is a popular API gateway we are proactively using. Sometimes we need to develop plugins to meet specific architecture/business needs.
第一次读这本书竟然已经是五年前了，最近在推上看到有人提及，就又回去翻了一下。还是有很多新感受（或者说记起来很多已经忘了的点），在这里梳理一下。
量化策略里有一大类叫做Pairs Trading-配对交易，是我觉得非常有趣的，其基本思路是寻找市场中表现相近的两个资产标的A和B, 当A涨得比较高的时候，卖出A买入B , 当A 跌的比较多的时候，买入A卖出B。这个思路很容易理解，因为如果两支股票确实相近，那么他们的价格有回归到相似的可能性，这也符合古老中国哲学里面“合久必分，分久必合”的朴素道理。
道理我们是都清楚的，甚至很多交易者也是模糊的按照这个思路来做的，比如当看到百事股价跟可口可乐股价偏离太远时，可能会考虑购进一些百事股票，卖出可口股票。问题是如何把这些模糊的想法进行量化分析，进而让算法捕捉到你看不到的机会，这主要需要解决两个问题：1. 如何发现符合要求的配对 2. 何时进场买卖 。 下面我们以QUANTAXIS为回测工具，以沪深市场为例来简要介绍下一个基本的配对交易策略是怎么进行的。
最近由于工作需要用到API网关做请求转发用，挑了目前人气较高的两款 Kong 和Goku 对比研究一下。