1
Retrieve
Answer from memory before opening docs. Failed recall marks what to study.
Train for the Tsuru Rust sample like a systems lab.
This is not a reading list. It is a retrieval-first, build-heavy learning system for Rust, binary parsing, packet formats, streaming reorder, market-data semantics, and performance discipline.
Target executable
Build a command-line parser that reads PCAP, extracts UDP quote payloads, parses market-data messages, optionally reorders within a 3-second window, and prints deterministic output.
PCAP->Ethernet->IPv4->UDP->Quote->Reorder
Next 90 minutes
- 1. Write what you remember about borrowed slices.
- 2. Build Lab 1 without opening docs for 20 minutes.
- 3. Read only the missing Rust Book sections.
- 4. Add three malformed-input tests.
- 5. Explain why the parser never panics.
Learning Loop
Each topic follows the same cognitive loop: recall first, read narrowly, build, break, measure, then explain.
2
Read narrow
Read only the slice needed for the next lab. No broad quant detours.
3
Build
Implement from bytes, errors, and tests. Prefer checked slicing over clever casts.
4
Break
Feed short packets, bad lengths, endian mistakes, and reordering edge cases.
5
Profile
After correctness, measure allocations and hot loops. Optimize measured paths.
6
Explain
Write one paragraph explaining the invariant, tradeoff, and failure mode.
Skill Map
The path is ordered by dependency, but practice stays interleaved: every phase includes correctness, malformed inputs, and explanation.
Days 1-7A fixed-width ASCII parser over &[u8] with no String allocation on the hot path.
Rust control of memory
Ownership is not syntax trivia; it is how packet bytes stay borrowed, bounded, and cheap.
Retrieval check
Explain why returning a borrowed field from a parser needs an input lifetime.
Common trap
Cloning into String to escape borrow errors.
Days 8-10parse-quote [-r] input.pcap skeleton with structured errors and golden CLI tests.
CLI and byte I/O
A market-data exercise is still a Unix tool: args, files, stdout, stderr, exit codes.
Retrieval check
Describe when BufReader helps and when reading one packet at a time still matters.
Common trap
Panicking on bad input instead of returning an actionable parse error.
Days 11-18PCAP -> Ethernet -> IPv4 -> UDP payload iterator with malformed packet tests.
Packet dissection
Headers are contracts. Every length, endian field, and offset must be checked before use.
Retrieval check
From memory: list fields needed to find the UDP payload inside IPv4.
Common trap
Assuming fixed IPv4 header length and ignoring IHL.
Days 19-23Quote-message parser plus stdout renderer matching the sample contract.
Quote semantics
Parsing bytes is not enough; the output must preserve timestamp, symbol, side, price, size, and order.
Retrieval check
Explain which fields are transport metadata and which fields are market data.
Common trap
Formatting early and losing numeric structure before tests can inspect it.
Days 24-30BinaryHeap reorder window that emits safe packets and flushes at EOF.
Streaming reorder
The 3-second buffer is a streaming algorithm, not a sort. Memory and latency are bounded.
Retrieval check
State the heap invariant and the condition that makes an item safe to emit.
Common trap
Collecting the whole input before sorting.
After first passCriterion benchmark, flamegraph note, and allocation reduction patch.
Performance discipline
Performance work starts after correctness and names the metric it moved.
Retrieval check
Name the exact benchmark, baseline, change, and result.
Common trap
Replacing readable code with clever code before profiling.
Diagnostic Router
Use symptoms, not vibes. Pick the route matching the thing that fails first.
Borrow checker fights every parser function.
Start with Lab 1 and explain lifetimes before reading packet docs.
Packet offsets feel like magic numbers.
Do Lab 2 and Lab 3 with one-byte truncation tests at every boundary.
Reorder sounds like sorting.
Skip broad algorithms. Learn heap invariant, then implement Lab 5.
Performance work becomes guessing.
Delay optimization until Lab 6: benchmark, profile, change one hot path.