Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
/**
* SQLite PRAGMA bootstrap for the conversation store.
*
* Applied once by the worker thread at boot, before any user query runs.
* See `.specs/mcp-hermes/implementation-plans/T2.1-sqlite-store.md` §6.1.
*/
import type { Database } from 'better-sqlite3'
export interface PragmaSetupResult {
walAvailable: boolean
journalMode: string
/** Effective `PRAGMA synchronous` level: 'FULL' (barrier mode) or 'NORMAL'. */
synchronous: 'FULL' | 'NORMAL'
}
export interface PragmaSetupOptions {
/**
* D3 (stateless-agents) — durability barrier. When true, `synchronous = FULL`
* (per-commit WAL fsync) — durable against node/power loss: every WAL commit
* fsyncs before the worker ACKs, so a host crash right after an acknowledged
* turn-boundary write can no longer lose it. Enabled by the stateless
* lifecycle (`CLERUM_STATELESS_LIFECYCLE=true`) or the explicit opt-in
* `CLERUM_DB_BARRIER_MODE=full`.
*
* Durability comes SOLELY from `synchronous=FULL`. The WAL checkpoint stays
* PASSIVE (WAL truncation only, NOT a durability mechanism) and its cadence
* (`CLERUM_DB_CHECKPOINT_EVERY_WRITES`, default 100) is intentionally
* unchanged — a per-turn PASSIVE checkpoint would add zero durability benefit.
*/
barrierMode?: boolean
}
/**
* Apply PRAGMAs and verify the WAL fallback path. Never throws — when the
* filesystem rejects WAL we log a warning and continue with the journal mode
* SQLite picked (typically `delete`).
*/
export function applyPragmas(db: Database, opts?: PragmaSetupOptions): PragmaSetupResult {
const synchronous: 'FULL' | 'NORMAL' = opts?.barrierMode === true ? 'FULL' : 'NORMAL'
// Order matters: synchronous + foreign_keys must apply after journal_mode.
db.pragma('journal_mode = WAL')
const journalMode = String(db.pragma('journal_mode', { simple: true }))
db.pragma(`synchronous = ${synchronous}`)
db.pragma('foreign_keys = ON')
// busy_timeout = 0 disables the internal back-off; the worker handles
// SQLITE_BUSY with a jittered custom retry (see busyRetry.ts).
db.pragma('busy_timeout = 0')
db.pragma('cache_size = -16000')
db.pragma('temp_store = MEMORY')
// mmap_size is best-effort — some sandboxes refuse and SQLite no-ops.
try {
db.pragma('mmap_size = 268435456')
} catch (err) {
console.warn('[SQLite] mmap_size pragma refused:', err)
}
return {
walAvailable: journalMode.toLowerCase() === 'wal',
journalMode,
synchronous,
}
}