Posts

TL;DR The optimal number of connections varies per workload and hardware. This benchmark observes how throughput, latency, and wait events change as connection count grows — it does not prescribe a universal number. $\text{core count} \times 2$ is a widely cited starting heuristic from the PostgreSQL wiki. Use it as a starting point, then tune upward until TPS plateaus and latency variance grows. 1. Problem Statement How many Postgres connections should I open? Should I open as many as possible? ...

Recently I learned that Postgres holds a lock on the rows it updates during a transaction and only releases it when the transaction finishes (commit/rollback). This surprised me because I always use select ... for update to lock the row before updating it in a transaction. I thought that other transactions could concurrently update the row that my current transaction is updating, causing a race condition. For example, I have this simple users table: ...

HDD Two important components that make up an HDD are a rapidly rotating platter and a movable magnetic head. The magnetic disk are divided into concentric circles called tracks. Each track is divided into equal smaller parts called sectors. Each sector is usually 512 bytes in size, and it is also the minimum unit of read and write operation. That means that even you only need to write or read a few bytes, the system must read the entire 512 bytes sector. To read or write any sector, the magnetic head must move to the track that contains the sector, and the platter must rotate to position the sector under the magnetic head. ...

To secure storing sensitive information, a common approach is to use envelope encryption: A master key (usually managed by a secure KMS) A row-level key (used to encrypt individual records) Each row of sensitive data (e.g., credit card information) is encrypted using a unique row key, the row key itself is encrypted using the master key and stored alongside the data in the database. This setup allows secure key rotation, better isolation, and compliance with security standards. ...

This post is part of the series Postgres for Everything. Nowadays, we typically develop stateless applications, making it easier to scale them horizontally. However, locking is an essential tool to prevent race conditions in software development. In applications with multiple instances, programming language-level locks are insufficient because they only work locally within an instance. A centralized locking mechanism, valid across all instances, is required. Redis SETNX You can use SETNX to implement a simple lock. SETNX sets a value for a key only if the key does not already exist. ...

From my previous post, I realized that pattern matching operators (LIKE, ILIKE) do not utilize indexes. As I explored further, I came across the concept of collation and decided to take some notes in this post. Encoding Encoding maps human-readable characters to numbers so computers can understand them. Essentially, it assigns a unique number to each character. Common encodings include UTF-8 and ASCII. ASCII: Represents 256 unique characters. UTF-8: Represents 1,112,064 characters, covering almost all characters from any language. Most modern programming languages, such as Go, natively support UTF-8. Unlike ASCII, which uses 1 byte per character, UTF-8 uses up to 4 bytes. Strings in programming languages are typically represented as byte arrays. In ASCII, the number of bytes corresponds to the number of characters. However, this is not true for UTF-8. ...

Have you ever asked, why do we usually listen on localhost:8080 during development process, can we listen on Google IP instead of localhost? I have a little code snippet written in NodeJS: 1 2 3 4 5 var http = require('http'); http.createServer(function (req, res) { res.writeHead(200, {'Content-Type': 'text/plain'}); res.end('Hello World\n'); }).listen(8080, 'localhost'); It doesn’t do anything complex, just starts a new server, listens on localhost:8080, and responds ‘Hello World’ to requests. I need you to do a little more thing, open your terminal, and execute this command (MacOS or Linux) to get your local IP: ...

If you are a Gopher, you probably heard some advice that goes like this: when you edit a value then use a pointer receiver, when you read use a value receiver. That’s not always correct and you should be careful when using them. Issue Today I encountered an issue which can be simplified by the bellow code snippet: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 package main import ( "fmt" "time" ) type Counter struct { Count int } // Method with pointer receiver func (c *Counter) Increment() { go func() { for i := 0; i < 10; i++ { c.Count++ time.Sleep(1 * time.Second) } }() } // Method with value receiver func (c Counter) Display() { go func() { for i := 0; i < 10; i++ { fmt.Println("counter value: ", c.Count) time.Sleep(1 * time.Second) } }() } func main() { c := Counter{Count: 0} c.Display() c.Increment() time.Sleep(15 * time.Second) fmt.Println("counter final value: ", c.Count) } I have a struct with two methods, Increment uses the pointer receiver to change the property of the struct, and Display uses the value receiver to read the value from the struct. Here is the result: ...

You may have heard about 0.1 + 0.2 != 0.3 already, it’s a common issue in most programming languages. The reason behind that is the floating point (IEEE 754) as the computer can’t represent exactly a decimal in binary. By utilizing the floating point, the computer can hold a very large decimal, but the trade-off is it now can only represent the approximate of the true value. However, in this post, I will not explain how floating point works but something else – constant. ...

this article assumes you have some basic knowledge about Rust smart pointers. the dereference (*) operator dereference operator * is used for getting the actual value from a reference: 1 2 3 let a: i32 = 5; let b: &i32 = &a; assert_eq!(*b, 5); in the above example, b is a reference to a, *b will return the value that b pointing to. the smart pointers Box<T> here is an example of how you can init and modify the value of a Box: ...