NKESSLING@TERMINALSCREEN WORKOpen to work from Nov 2026RTM 17:28:39 · UTC 15:28
>
WORK ~/work

Storage engines, build tools and queues. Every project shows the problem, the file that fixed it and the numbers after.

PROJECTS
8
STARS
11.7K
LANGUAGES
Rust · Go · TS · Zig
SINCE
2017
SCRN WORK · SCREENER8 OF 8 SHOWN

All projects, screened.

Filter by language, sort any column, and hover a row for the preview.

F3 REPOS · 4 PROJECTS
nkessling / tidewater⑂ mainRust★ 4,812⑂ 211last commit 2h ago
README.mdrendered

# tidewater

Crash-safe write-ahead log for embedded databases.

buildpassingcrate2.1.0msrv1.76licenseMIT

## Why

Group commit batches concurrent writers into one fsync. After a segment rotates, the directory is fsynced too, so a crash can never leave a segment the log cannot find.

## Install

$ cargo add tidewater

## Numbers

workloadtidewaterfsync each write
append 4 KiB1.92 GB/s0.21 GB/s
p99 commit2.1 ms9.8 ms
recovery 8 GB3.4 s11.0 s

Runs under three production databases and one payments ledger.

COMMITS · main8 shown
  1. fix(wal): fsync the directory after a segment rotates
    a41f9c2 · 2h ago
  2. Merge #412: group commit for concurrent writers
    9e03b71 · 1d ago
  3. group: wake all waiters on one fsync
    c7d2a18 · 1d ago · feature branch
  4. group: batch writers behind a ticket lock
    51b6e40 · 2d ago · feature branch
  5. bench: 1.92 GB/s sequential append on M3
    0f9a3d5 · 3d ago
  6. test: 40k simulated power cuts, 0 lost writes
    e2c84b9 · 4d ago
  7. release v2.1.0v2.1.0
    7b19f06 · 6d ago
  8. refactor: split recovery into scan and replay
    3ad50ce · 8d ago
ACTIVITY · 26 WEEKS · TIDE
README · scroll to continue8 files · 8 recent commits · crate 2.1.0
BENCH BEFORE / AFTER01 / 04

Before and after, measured.

Each project has one number that mattered. Scroll to draw the after line over the before line.

  1. TIDEAppend throughput by record size9.1x
  2. KILNMedian CI time, 30 days6.1x
  3. LOOMQuiet tenant p95 pickup during a burst1,079x
  4. PRSMDiff time by trace size1,166x
TIDE · APPEND THROUGHPUT BY RECORD SIZE
9.1xmore throughput
BEFORE 0.26 GB/sAFTER 1.92 GB/s
00.5381.081.612.15
record size, 256 B → 64 KiBfsync each write vs group commit, M3, one core
RATE CLIENTS · ANALYST NOTES6 NOTES

What clients wrote afterwards.

Filed like analyst notes: the rating, the result it was based on, and who wrote it.

CONSENSUS6 OF 6 · HIRE AGAIN
0 HOLD · 0 PASS
RATING · HIRE AGAIN0 lost writes since
“Noa found the durability bug we had been blaming on hardware for a year, fixed it, and left us a crash test so it stays fixed.”
Priya RamanCTO · Vantle Ledger
RATING · HIRE AGAINCI 18 m → 3 m
“Our CI went from a coffee break to a sip. The sizing guide alone saved us a quarter of platform work.”
Tomasz WieczorekHead of Platform · Harbourmeal
RATING · HIRE AGAINHandover in 2 days
“Clear writing, clear pull requests, and no surprises. The handover doc read like a good README.”
Amara OkaforEngineering Manager · Norrvik Pay
RATING · HIRE AGAINp95 41 s → 38 ms
“The fair queue change was eight lines and it ended a year of support tickets about slow emails.”
Jonas BrekkeStaff Engineer · Postquill
RATING · HIRE AGAIN−48% code shipped
“Noa asks the question that makes the design smaller. We shipped half the code we planned and it was faster.”
Lucía FerreyraVP Engineering · Tallyroom
RATING · HIRE AGAIN11 → 1 pages/week
“Pages went from eleven a week to one. The team finally took real holidays.”
Daniel OseiSRE Lead · Quaymark Health
EOF END OF PAGE · NKESSLING↑ TOP

Type CONTACT, or just write.

hello@example.com
>

STATUS

Open to work from Nov 2026

Replies within one working day

RTM 17:28:39 · UTC 15:28

COLOPHON

Set in Geist, Geist Mono and Doto.

Charts drawn from commit data, live or demo.

No trackers, no cookies.

Create a free website with Framer, the website builder loved by startups, designers and agencies.