No matching posts.
Deservewhatyouwant.
All Articles
CI queues, shared Mac runners, and the deployment contention that remained after distributing image builds.
I tested restoration, application behavior, and worker isolation with production data before migrating NCP accounts. The rehearsal exposed extension schema, connection target, and leftover Redis key issues, while the actual cutover verified matching data and about 13 minutes of user impact.
I compared three ways to migrate PostgreSQL and Redis between NCP accounts. Data size and operating conditions favored a short pause in writes and dump-and-restore, with the resumption of writes defining the cutover boundary.
Before moving production to a new NCP account, I reconciled the infrastructure documentation with the resources actually running. I rebuilt Terraform, NKS, and the deployment path, then moved the data and traffic with about 13 minutes of user impact.
I traced similar sources across different topics back to the actual search queries and code. Grok expresses the question, while code preserves search intent and assigns platform sources.
Letting Grok judge meaning while code preserves and validates titles
How beta users' questions and behavior changed Wiblog's customer hypothesis, onboarding, B2C/B2B direction, and value proposition.
Wiblog collected references for topics users wanted to write about, then generated and published source-backed posts. This post follows the path through custom-domain connection and Google Search visibility, including its limits.
Separate workspaces and permissions, with one shared API tunnel
Detect, escape, fall back — for drop-offs no log catches
From a Grafana 88% alert to automatic failover at 80%
Separating immediate setup from work that waits for DNS, with connection checks and registration retries
Recomputing 15 topics from logs: ₩309 down to ₩150
Why fast repeated Grok calls beat the better writer
How indexing observations changed Wiblog's domain recommendations
Replacing two LLM calls with one autocomplete request
Removing 60–100s outages, then the last 5 seconds
42.8% errors in 9 minutes, caused by one updateMany
A 7-node production clone, rebuilt with one make command
Five years of hang issues behind a perfect-looking active-replica
Surviving a 300MB heap and tab crashes with 5-minute savepoints
