DEV Community

jidonglab profile picture

jidonglab

1 project a week. Building and sharing the entire process — from idea to shipped product in 7 days. Currently: AI news automation.

Nginx proxy_buffering Ate My LLM Stream: 0.4s First Token Became 6.1s

Nginx proxy_buffering Ate My LLM Stream: 0.4s First Token Became 6.1s

1
Comments
7 min read

Want to connect with jidonglab?

Create an account to connect with jidonglab. You can also sign in below to proceed if you already have an account.

Already have an account? Sign in
temperature=0 Isn't Deterministic: 17 of 5,000 Greedy Tokens Flipped

temperature=0 Isn't Deterministic: 17 of 5,000 Greedy Tokens Flipped

1
Comments
7 min read
docker stop Takes 10 Seconds: 5 of My 6 Containers Ignored SIGTERM

docker stop Takes 10 Seconds: 5 of My 6 Containers Ignored SIGTERM

1
Comments 5
6 min read
asyncio.gather Kept Running: 1,761 LLM Calls Finished After My 500

asyncio.gather Kept Running: 1,761 LLM Calls Finished After My 500

Comments
7 min read
Refmade 0.4 Is Out: A Free Mac and Windows App in Eight Languages

Refmade 0.4 Is Out: A Free Mac and Windows App in Eight Languages

Comments
7 min read
Claude Prompt Caching Minimum: 1,240 Haiku 4.5 Calls, 0 Cache Hits

Claude Prompt Caching Minimum: 1,240 Haiku 4.5 Calls, 0 Cache Hits

1
Comments 1
6 min read
KV Cache Math: Llama 3.1 8B Needs 16 GiB Just to Hold 128K Tokens

KV Cache Math: Llama 3.1 8B Needs 16 GiB Just to Hold 128K Tokens

1
Comments
6 min read
GitHub Actions fetch-depth: 1 Turned 9 Releases Into 0.1.dev1

GitHub Actions fetch-depth: 1 Turned 9 Releases Into 0.1.dev1

Comments
6 min read
SQLite 'database is locked' in 2ms: Why busy_timeout Didn't Help

SQLite 'database is locked' in 2ms: Why busy_timeout Didn't Help

Comments
6 min read
vLLM Prefix Caching: One Timestamp Dropped My Hit Rate From 89% to 3%

vLLM Prefix Caching: One Timestamp Dropped My Hit Rate From 89% to 3%

Comments
7 min read
GitHub Actions Cache Hit, Never Saved: 312 Runs on a 47-Day-Old Cache

GitHub Actions Cache Hit, Never Saved: 312 Runs on a 47-Day-Old Cache

Comments
7 min read
Eight Languages in One Day: Localizing a Korean-Only App

Eight Languages in One Day: Localizing a Korean-Only App

Comments
6 min read
Whisper Hallucinates on Silence: 61 of 700 Pauses Became 'Thank You'

Whisper Hallucinates on Silence: 61 of 700 Pauses Became 'Thank You'

Comments
6 min read
Claude API max_tokens Cut Off 143 of 1,600 Summaries Mid-Sentence

Claude API max_tokens Cut Off 143 of 1,600 Summaries Mid-Sentence

Comments
6 min read
all-MiniLM-L6-v2 Silently Truncates at 256 Tokens: 38% of My Chunks

all-MiniLM-L6-v2 Silently Truncates at 256 Tokens: 38% of My Chunks

Comments
6 min read
Docker Compose depends_on Didn't Wait: 11 of 40 Fresh Starts Crashed

Docker Compose depends_on Didn't Wait: 11 of 40 Fresh Starts Crashed

Comments
6 min read
5 Releases in One Day: Release Gates for a Solo App

5 Releases in One Day: Release Gates for a Solo App

Comments
6 min read
Claude Code PreToolUse Hook: 80 of 91 Secret Blocks Were False Alarms

Claude Code PreToolUse Hook: 80 of 91 Secret Blocks Were False Alarms

1
Comments 3
6 min read
Claude Code on zsh: 7,795 Commands Broke, 5,559 Still Exited 0

Claude Code on zsh: 7,795 Commands Broke, 5,559 Still Exited 0

Comments
6 min read
LangChain chunk_overlap=200: 2,181 of 4,155 Chunks Got No Overlap

LangChain chunk_overlap=200: 2,181 of 4,155 Chunks Got No Overlap

Comments
6 min read
Refmade Shipped Without Git: 2,764 Files in One Snapshot

Refmade Shipped Without Git: 2,764 Files in One Snapshot

Comments
5 min read
Claude Code Env Vars Don't Persist: 3 of 80 Runs Fell Back to Prod

Claude Code Env Vars Don't Persist: 3 of 80 Runs Fell Back to Prod

1
Comments
6 min read
Claude Code Token Usage: 97% of 44.6B Tokens Were Cache Re-Reads

Claude Code Token Usage: 97% of 44.6B Tokens Were Cache Re-Reads

1
Comments 2
6 min read
RoPE in bfloat16: 8,192 Token Positions Collapsed Into 897

RoPE in bfloat16: 8,192 Token Positions Collapsed Into 897

Comments
6 min read
Shipping a Mac Electron App Without Apple Notarization

Shipping a Mac Electron App Without Apple Notarization

5
Comments
5 min read
json.dumps Writes NaN by Default: 31 LLM Reports JSON.parse Rejected

json.dumps Writes NaN by Default: 31 LLM Reports JSON.parse Rejected

Comments
6 min read
asyncpg + PgBouncer Transaction Mode: 61 Prepared Statement Errors

asyncpg + PgBouncer Transaction Mode: 61 Prepared Statement Errors

Comments
6 min read
GitHub Actions GITHUB_TOKEN Doesn't Trigger Workflows: 11 Tags, 0 Runs

GitHub Actions GITHUB_TOKEN Doesn't Trigger Workflows: 11 Tags, 0 Runs

Comments
6 min read
Python requests timeout=60 Let My LLM Stream Run for 38 Minutes

Python requests timeout=60 Let My LLM Stream Run for 38 Minutes

1
Comments 3
7 min read
My Claude Code Sign-in Fix Needed Two Fixes of Its Own

My Claude Code Sign-in Fix Needed Two Fixes of Its Own

Comments
5 min read
LLM-as-Judge Position Bias: Swapping A/B Flipped 87 of 300 Verdicts

LLM-as-Judge Position Bias: Swapping A/B Flipped 87 of 300 Verdicts

Comments
7 min read
git push --force-with-lease Is Unsafe If Your Editor Auto-Fetches

git push --force-with-lease Is Unsafe If Your Editor Auto-Fetches

1
Comments
6 min read
Python ThreadPoolExecutor Swallowed 212 Claude API Errors Silently

Python ThreadPoolExecutor Swallowed 212 Claude API Errors Silently

Comments
6 min read
AWS ALB Idle Timeout Killed 14% of My Voice WebSockets at 60s

AWS ALB Idle Timeout Killed 14% of My Voice WebSockets at 60s

Comments 2
6 min read
Measuring What My Electron App Sends: 94 Sockets, 0 Out

Measuring What My Electron App Sends: 94 Sockets, 0 Out

1
Comments 1
5 min read
vLLM Speculative Decoding: 1.8x Faster Alone, Slower at 32 Users

vLLM Speculative Decoding: 1.8x Faster Alone, Slower at 32 Users

2
Comments 1
7 min read
Claude Code Deny Rules Block Read(.env) But Not cat .env

Claude Code Deny Rules Block Read(.env) But Not cat .env

2
Comments 4
6 min read
pgvector HNSW Filtering Returned 0 Rows for 52 of My 400 Queries

pgvector HNSW Filtering Returned 0 Rows for 52 of My 400 Queries

Comments
6 min read
frequency_penalty Broke My JSON: 47 of 88 Long Invoices Lost Items

frequency_penalty Broke My JSON: 47 of 88 Long Invoices Lost Items

Comments
7 min read
Wrapping Claude Code and Codex in Electron: A Stop Bug

Wrapping Claude Code and Codex in Electron: A Stop Bug

Comments
6 min read
Docker Bypasses UFW: 3 Ports I Denied Were Open to the Internet

Docker Bypasses UFW: 3 Ports I Denied Were Open to the Internet

Comments
6 min read
Silero VAD Endpointing Cut Off 23% of Answers: 700ms Isn't a Turn

Silero VAD Endpointing Cut Off 23% of Answers: 700ms Isn't a Turn

Comments
6 min read
nginx proxy_buffering Broke My Claude SSE Stream: 0.7s to 3.4s TTFT

nginx proxy_buffering Broke My Claude SSE Stream: 0.7s to 3.4s TTFT

2
Comments 1
7 min read
BPE Tokenization: A Trailing Space in My Prompt Broke 41 of 300 Labels

BPE Tokenization: A Trailing Space in My Prompt Broke 41 of 300 Labels

Comments
6 min read
13 Research Agents Told Me My Feature Was Already Free

13 Research Agents Told Me My Feature Was Already Free

Comments
5 min read
GitHub Actions paths Filter + Required Checks: 23 PRs Stuck Pending

GitHub Actions paths Filter + Required Checks: 23 PRs Stuck Pending

Comments
6 min read
Celery visibility_timeout Ran My Claude Jobs 3.3x: 1,800 Became 5,913

Celery visibility_timeout Ran My Claude Jobs 3.3x: 1,800 Became 5,913

Comments
7 min read
JSON Schema Field Order: My vLLM Classifier Answered Before Thinking

JSON Schema Field Order: My vLLM Classifier Answered Before Thinking

Comments
6 min read
Claude Prompt Caching Hit 0 of 1,412 Calls: A Timestamp Killed It

Claude Prompt Caching Hit 0 of 1,412 Calls: A Timestamp Killed It

Comments
7 min read
Codex Image Gen Saved My White Pixels as Transparent

Codex Image Gen Saved My White Pixels as Transparent

Comments
6 min read
GitHub Actions Cache Never Updates: 9 Weeks of a Frozen mypy Cache

GitHub Actions Cache Never Updates: 9 Weeks of a Frozen mypy Cache

Comments
6 min read
Claude Batch API Returned Results Out of Order: 59% of My Labels Wrong

Claude Batch API Returned Results Out of Order: 59% of My Labels Wrong

Comments
6 min read
vLLM KV Cache Math: Why My 24GB GPU Served Only 4 Requests at Once

vLLM KV Cache Math: Why My 24GB GPU Served Only 4 Requests at Once

Comments
7 min read
Docker RUN rm Didn't Delete My .env. 3 of 14 Images Still Had It

Docker RUN rm Didn't Delete My .env. 3 of 14 Images Still Had It

Comments
6 min read
I Shipped a Friends Feature, Then Dropped It the Same Day

I Shipped a Friends Feature, Then Dropped It the Same Day

Comments 1
5 min read
SQLite 'database is locked' Ignored My 30s busy_timeout: 311 Failures

SQLite 'database is locked' Ignored My 30s busy_timeout: 311 Failures

Comments 1
6 min read
Claude stop_reason max_tokens: 83 Reports Silently Cut Off Mid-JSON

Claude stop_reason max_tokens: 83 Reports Silently Cut Off Mid-JSON

Comments
6 min read
Five Parallel AI Builders, One Contract, 25 Review Fixes

Five Parallel AI Builders, One Contract, 25 Review Fixes

Comments 2
6 min read
Refmade: a free desktop app for Claude Code and Codex, without the terminal

Refmade: a free desktop app for Claude Code and Codex, without the terminal

1
Comments
3 min read
BM25 vs Embeddings: Why My Vector Search Couldn't Find Error E4012

BM25 vs Embeddings: Why My Vector Search Couldn't Find Error E4012

1
Comments 1
7 min read
loading...