Understanding Log - Part 1.
I was studying the Write-Ahead-Log for a personal project and decided to do a presentation at DULSS. I came up with the slides and thought I was done, until I got a feedback from Tushar and Chiradip that pushed me to go deeper. This push led me to see Log in several systems I didn't know before.
Think WAL as a state machine, model in that way, and apply state machine in applicable scenarios like storage.
A deterministic state machine drives its execution through a Write-Ahead Log (WAL), which forces all incoming state transitions to be durably ordered before they alter the application state. Because this linear ledger acts as the authoritative timeline, the state machine can infallibly reconstruct its exact, consistent state during recovery simply by sequentially replaying the logged inputs. — Chiradip's feedback
In this series we will be looking at the Log in several systems
- in databases
- in version control(think
git log😉) - in distributed systems (think
kafka and its partitions) - in consensus algorithms (think
raft and vsr)
Before we go into each of these, let's understand what a Log is first.
Overview
A Log is actually just a state machine tracker. It is a system that tracks the changes to the particular state, it tracks
"what altered a state" and "when a state was changed" (in an orderly manner). Look at a log as a Table with entries
appended from Left to Right and sorted by time .It builds on the principle of State Machine Replication
If two deterministic processes starting from the same initial state are fed with the same input in the same order then these processes will end up in the same state and same output—State Machine Replication
The two key things in this principle is same input and same order take note.
Purposes of a log
A log basically serves as
- Authoritative Source
- State Machine Tracer
Log as an authority.
You might have had an experience where an issue happened in production, and you had to inspect it; the first thing that comes to mind most times is to check the "log." Now logs in this scenario can mean many things for different developers/cases for some, It means an output to the console or a file or a system. However, irrespective of the form it takes, the log serves as an "authority" to know/trace/track what happened (what caused the issue) and when the issue happened. In this series we are looking at programmable logs, not just 'logs' from application consoles.
Log as a state machine
If we say a log tracks state-altering actions, it means that we can feed these tracked-actions into a system at the initial state
and arrive in the finished state. It just reminds me of finite-state-machines in compilers and d-tran function that allows
a state to transition to another on an input.
Properties of a Log
- Insertion: A log can be inserted to from left-to-right only or top-bottom only. You cannot append a particular position in a log only the end.
- Segmentation: Once we start appending to a log, the size begins to grow, and it because too large for a process, a log can now segment itself into smaller chunks of logs that allows faster processing.
- Durability: This is one key area of a log, it must be durable, the contents of the Log should not be lost in case of a crash. If a log is not durable, then it cannot act an authoritative source.
Shape of a Log
We discussed the purposes and properties of a log, now let's look at the shape of a log in an abstract form not diving into how a log looks like in a database, kafka or raft, just an overview to give you a mental model of what we will be studying through these series.
A log can be seen as an array of entries that are appended from left-to-right and are sorted by identifiers. We all know arrays; they're linear data-structures where values are accessed by their index. Good! Imagine an array that arranges all its inputs in an orderly manner, you can't put at index 10 (i[10]) when you've not filled index 9 (i[9]), that is simply a log.
Conclusion
From this I hope we can now envision the ride we are about to take. This introduction gives an eagle-view of a log; we are now
going to deep dive into a log in several systems like databases and how logs helps in crash recovery, how version control systems
use logs for history and versioning, how consensus algorithms like view-stamped replication are basically distributed logs, how kafka partitions
are also durable log streams.
Something you keep in your mind
Many concepts in computer science/software technology replay themselves in different systems using different names