docs/generated/vocabulary.mdgenerated from code
Reference · 3 min read
Constants: budget
Vocabulary: Constants in budget.
Read as
Source: crates/lvcs-vocab/src/budget.rs.
| Name | Value | Meaning |
|---|---|---|
MACHINE_RSS_BYTES | 2 * 1024 * 1024 * 1024 | The most resident memory a machine's LVCS processes hold together, the daemon and every plugin, at their peak, whatever the number and size of the projects: 2 GiB (the owner's ruling for 0.3.0, revising the 1 GiB of discussion:owner/632 and discussion:owner/790). More projects may cost speed, never memory. Everything sized from the memory cap derives from this. |
PLUGINS_RSS_BYTES | 256 * 1024 * 1024 | The plugins' share of [MACHINE_RSS_BYTES], every plugin process together, its plugin commands included: 256 MiB (the orchestrator's ruling on discussion:owner/790). |
PLUGINS_RSS_FLOOR_BYTES | 128 * 1024 * 1024 | The least of [MACHINE_RSS_BYTES] the plugins are left whatever the daemon holds: 128 MiB (discussion:owner/790). |
DAEMON_IDLE_RSS_BYTES | MACHINE_RSS_BYTES - PLUGINS_RSS_BYTES | The most resident memory an agentflare daemon holds at its peak, at rest, opening, importing, and in normal work, whatever the number and size of its projects, in bytes: what [MACHINE_RSS_BYTES] leaves beside the plugins' share, 1,792 MiB (discussion:owner/790's rule, at the 2 GiB cap). Its rester's resident mark and every transient it admits are sized from it. |
DAEMON_IDLE_CPU_PERMILLE | 10 | The most CPU an idle agentflare daemon uses, in thousandths of one core: 10, which is 1%. |
PLUGIN_IDLE_RSS_BYTES | 100 * 1024 * 1024 | The most resident memory an idle plugin process holds, in bytes: 100 MiB. |
DAEMON_LOG_BYTES_PER_MIN | 64 * 1024 | The most an agentflare daemon logs per minute of its life at the default LVCS_LOG filter, in bytes of log text (each record's target, message, and fields as key=value, its stderr line less the timestamp): 64 KiB. Sampled e2e daemons (2026-11-07, read mid-scenario: ci_channel, channels_forward, rewrite_save, grant_revoke) logged 1.1 to 5.8 KiB a minute, so this is ten times the busiest; the maintainers' daemon that filled 38 GB (e2e-suites/3) logged 10 to 40 MiB a minute, one warning repeated, which is hundreds of times over. A process that lived under a minute is measured over one. |
PLUGIN_LOG_BYTES_PER_MIN | 16 * 1024 | The most a plugin process logs per minute of its life, measured as [DAEMON_LOG_BYTES_PER_MIN]: 16 KiB. Sampled e2e plugin processes (actions, secrets) logged 0.2 to 1.2 KiB a minute, so this is over ten times the busiest. |
PROJECT_STATE_GROWTH_BYTES | 64 * 1024 * 1024 | The most a node's state grows over one e2e scenario, in bytes: its project's .lvcs/ and its daemon's state directory, from an empty volume at the container's start to the daemon's stop: 64 MiB, one journal segment (limits::JOURNAL_ROTATE_BYTES). Sampled nodes (2026-11-07, read late in their scenarios) held 0.3 to 4.9 MiB, nearly all of it the journal, so this is thirteen times the largest. |