Kravtsov Mykhailo

A server monitoring service for Linux. It reads web server, auth, application and auditd logs, watches files, listening ports, processes and their outbound connections and the database, mirrors requests, sets traps and canary tokens, and matches injection signatures. What it notices is reported to a file, to Telegram, to an HTTP receiver or by e-mail, and the log can be shipped to another server. It changes nothing in the system until it is asked to.

Events are chained into incidents, actors are scored across sources, normal behaviour is learned so that noise is suppressed and thresholds can be derived from the server's own traffic; blocking and response actions are available but opt-in, and the service protects itself as well. A single Go executable with no external dependencies beyond the database drivers, configured by one file, installed as a systemd service and tied to no particular application: everything server-specific lives in the configuration.

A universal Redis Streams → PostgreSQL pipeline in Go with a configurable mapping: one program handles many kinds of data, because which stream to read and which table and columns to write it to is described in a mapping file, with no code changes. Several Redis sources feed one Postgres store; connections are made over mTLS and can go through a proxy list with failover. Parallel workers read each stream through consumer groups, convert fields by the declared SQL types and write one multi-row insert per batch with ON CONFLICT handling; tables can be created automatically from the mapping.

Reliability is explicit: at-least-once delivery, where a message leaves Redis only after it is durably stored in its table or in the dead-letter table; transient failures such as a network drop or an unavailable database keep the batch in Redis for a retry, while only genuinely broken data is dead-lettered; circuit breakers per source and for Postgres; exponential backoff with jitter; reclaim of messages stuck behind a crashed worker; graceful shutdown that drains and loses nothing.

Runs as a hardened systemd service with watchdog and journal logging, exposes an admin HTTP endpoint with metrics, rate-limits polling per source, cleans up dead letters by TTL and keeps secrets and the schema out of the repository.

A Node.js client for reading and managing Java applications through JMX, targeting Node.js 24+ and the Java 25 toolchain. It runs a small Java sidecar and exchanges newline-delimited JSON with it over the child process's standard streams, while the sidecar speaks the standard JMX remote protocol (JSR-160, RMI by default) to the target. Nothing is deployed into the monitored JVM: it only needs an enabled JMX remote connector, which makes the package the right choice when the target deployment cannot be changed.

The API covers attribute reads and writes, batch attribute reads, MBean listing by pattern, domain and MBean counts, operation invocation with explicit signatures, and JMX notification subscriptions delivered as events. Every operation returns a Promise and still honours the historical callback form. Connections accept a host and port or a full JMX service URL with credentials, an optional local classpath for application-specific classes, per-request timeouts and a debug hook that receives every protocol message.

Java values are decoded carefully: longs outside the safe integer range keep their exactness, CompositeData and TabularData become typed wrappers and row arrays, Maps with string keys become plain objects, and Java errors surface with their class name and stack trace attached. The client guards against a misbehaving sidecar: stdout is reserved for the protocol, stderr is capped and attached to errors, and a flood of diagnostic output terminates the sidecar with all pending requests rejected.

Published on npm with the built sidecar JAR included, tested on every push through GitHub Actions, and benchmarked at about 14,000 requests per second with sub-millisecond p99 latency inside a container and roughly 23,000 requests per second natively. Credentials travel only through the local process pipe and never reach logs.

Image checkboxes that behave like a real <input type="checkbox">: any PNG, SVG, WebP, GIF or AVIF becomes a toggle, shown in full colour when checked and in grayscale, or any filter you choose, when not. One self-contained 30 KB file, zero runtime dependencies, no build step, usable from a module script tag or imported into React, Vue, Angular or Svelte.

The element is a form-associated custom element: FormData sees its value, required validation shows the native browser bubble, form.reset() and a disabled fieldset work, and back/forward navigation restores state; browsers without ElementInternals fall back to a hidden input. The API mirrors HTMLInputElement: checked, indeterminate, disabled, required, validity, labels, checkValidity and reportValidity. Label position and size are configurable, extra attributes stay on the host element, styling goes through custom properties and data attributes, and the package is tested in CI and published with a live demo.

A zero-dependency vanilla-JS data table for dashboards: no jQuery, no Bootstrap, multiple independent instances per page, loaded from npm or a CDN. Search, sort and pagination run either on the client or on the server through a fetch callback. Thirteen column types cover text, numbers, links, copy buttons, status badges, selects and tags, images, textareas, downloads, custom HTML and row actions.

Rows can be edited inline with a save callback, expanded into accordion details, selected with bulk actions, and removed through a built-in confirmation dialog. Columns can be picked, reordered and filtered one by one. Export to CSV, XLSX and PDF through pluggable adapters. Large sets use virtualization that renders only the rows in view; narrow screens switch to a card layout. Loading, error and skeleton states, grid keyboard navigation with ARIA attributes, translatable labels, theming through CSS variables with dark and light modes, TypeScript types and wrappers for React, Vue, Angular and Svelte.

An image drop zone with previews: a dependency-free front-end widget (16 KB of JS and 2 KB of CSS gzipped) and a framework-free Node upload handler. Files arrive by drag and drop, the file picker, pasting or the camera; each one appears as a thumbnail drawn from its own bytes, with name and size, decoded one at a time so a large selection shows its first picture at once. The queue can be reordered, and limits apply per file, per queue, per file count and per decoded pixels, so a tiny PNG declaring a huge canvas cannot exhaust memory.

Without an endpoint it is just a form field and the files ride along with the form. Given an endpoint it uploads on its own, with progress bars, a working cancel button, retries only for connection failures, and mass uploads that survive an interruption. The server half is a single request handler that mounts in Express, AdonisJS, Fastify, Nest or bare node:http with a multipart parser as its only dependency: it validates files by their leading bytes rather than their names, refuses unsafe names instead of repairing them, writes each upload atomically and rejects cross-origin posts.

Optional blocks switch on only when configured: shrinking pictures on the page before they are sent (resize to a box, re-encode at a chosen quality or strip metadata losslessly), per-session folders with an owner on every answer, malware screening through a hash lookup or a scanner of your own where only the hash leaves the server, and server-side upload speed limits. Five languages, light and dark themes with every colour as a CSS variable.

An embeddable file manager: a dependency-free front-end widget and a framework-free Node handler over a real filesystem. The back end is a single request handler that mounts in Express, AdonisJS, Fastify, Nest or bare node:http, confines the client to one root directory, enforces operation permissions on the server side and exposes an authorize hook and a REST API. The widget mounts into any DOM element and works with any front-end framework or none.

Beyond the eight basic operations it offers a lazily loaded folder tree, tiles and table views, drag and drop including uploads from the OS, keyboard navigation, image thumbnails and a viewer for images and PDF, packing and unpacking of 13 archive formats with safe extraction, search across the tree by masks, regular expressions and file contents in a worker thread, a built-in terminal with its own commands, editors for spreadsheets (csv, xlsx, xls, ods), documents (docx, odt, doc) and code with highlighting, PDF page operations, five interface languages, light and dark themes and respect for prefers-reduced-motion.