All articles tagged with Linux
Building Prometheus from source on a host with older Go and Node than the repo asks for: what the build does, what warns, and how I verified the result.
Dokku's nginx templates listen on [::] by default. Here is why nginx refuses to start on a host without IPv6, and how to fix it, including notes forβ¦
How to set up a single-node edge MQTT aggregator with K3s and WireGuard, and how to fix the faulty flags, units and sysctls that edge guides often repeat.
A Bun or Node.js service can be slow on a VPS because of CPU steal or slow disks. How to measure both, fix the host first, and then compare runtimes fairly.
How to link a hyperscaler workload to a database on a Norwegian VPS with WireGuard, what the tunnel does not fix, and the common mistakes in such setups.
Why Kubernetes, Docker Swarm and Nomad clusters turn flaky on slow disks or stolen CPU, how to measure it with fio and vmstat, and what to change.
How to remove common latency and connection-error causes in an Nginx or Kong API gateway: file limits, upstream keepalive, backlog, port range, logging and TLS.
How to run a three-node NATS JetStream cluster (server 2.6, nats.go 1.13) with replicated streams, authentication, correct acks and sane OS limits.
How to measure request latency, CPU steal and disk trends on your own server with Nginx timing logs, Prometheus 2.28 and Grafana 8, without a third-party APM.
Replace trust-by-network-location with encrypted, authenticated links: WireGuard between nodes, nginx client certificates and hardened SSH with a second factor.
Slow or timing-out Jenkins builds are often caused by iowait, swapping and a badly sized JVM heap. How to diagnose and tune each on a 2014-era Linux server.
How to keep a local HAProxy in step with a ZooKeeper service registry: ephemeral znodes, a watcher that regenerates the config, and a graceful reload.
When to use Memcached or Redis 2.8 for caching and PHP sessions on CentOS 6. Covers persistence, eviction, AOF fsync stalls, and a split-instance setup.
How to keep slow jobs out of the web request path with RabbitMQ, Celery 3.0 and Nginx 1.2 on Debian 7, with corrected configuration and the limits of the setup.
How to build reports from raw Apache or Nginx access logs with AWStats on CentOS 6: log format, config, cron, protected access, and limits.
A reference setup for Postfix on CentOS 6 (2012): fix mail rejected or sent to spam with matching DNS, restricted SMTP, TLS submission with SASL, SPF and DKIM.
Put HAProxy 1.4 in front of web nodes on CentOS 6, tune the kernel and use Keepalived VRRP so the balancer itself is no single point of failure.
How to tell when an OpenVZ container is being refused resources: read user_beancounters, interpret failcnt, and shrink MySQL and Apache to fit the limits.
Build an OpenVPN 2.2 gateway on CentOS 6 with a private CA, and fix the usual failures: missing TUN device, no forwarding or NAT, and iptables rule order.
How to confirm that MySQL 5.5 is stuck on disk I/O, and how to tune the I/O scheduler, ext4 mount options and InnoDB settings for flash or fast virtual storage.
How to run a private, key-only Git server with Gitolite 2.x on CentOS 6, with access rules kept in a Git repository. Covers latency, I/O, and backups.
Cut disk waits on a MySQL 5.5 server on CentOS 6: convert tables to InnoDB, size the buffer pool and logs, set the flush policy, I/O scheduler and swappiness.
Why Apache prefork wastes RAM on static files and slow clients, and how a lighttpd front end plus tuned MaxClients and keep-alive fixes it on CentOS 5.
A 2009 reference for choosing between a VPS and a dedicated server. It covers how to find and fix OpenVZ memory-limit failures with user_beancounters.
Why services in an OpenVZ container fail with allocation errors while free memory looks fine, and how to find and raise the exhausted User Beancounter limit.
How to remove the single points of failure from a small web cluster in 2009: two HAProxy nodes sharing a VRRP address in front of two or more web nodes.