feat(httpapi): harden control plane and speed up prefix refresh

Очередь job получила lease и fencing, телеметрия и прогресс deploy сохраняются, ревизия пишет manifest входов. Обновление префиксов не переписывает неизменённый снимок, большие наборы идут через COPY, а устаревшая ревизия не попадает в deploy. Бандл ноды кэшируется, prune и постановка job больше не держат горячий путь.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
DenozordecandCursor committed 2026-11-02 17:23:59 +07:00
1 parent 924f111e48
commit f8e0520817
197 files changed
+18655 -1416

No files matched your search

@@ -0,0 +1,12 @@
DROP INDEX IF EXISTS idx_job_audit_expired_lease;
DROP INDEX IF EXISTS idx_job_audit_claim;
ALTER TABLE job_audit DROP COLUMN request_sequence;
ALTER TABLE job_audit DROP COLUMN cancel_requested;
ALTER TABLE job_audit DROP COLUMN available_at;
ALTER TABLE job_audit DROP COLUMN fencing_token;
ALTER TABLE job_audit DROP COLUMN attempt;
ALTER TABLE job_audit DROP COLUMN lease_until;
ALTER TABLE job_audit DROP COLUMN owner_id;
ALTER TABLE job_audit DROP COLUMN request_hash;
ALTER TABLE job_audit DROP COLUMN request_json;
DELETE FROM tenant WHERE id = '00000000-0000-0000-0000-000000000000' AND slug = 'evobgp-system-jobs';
@@ -0,0 +1,14 @@
ALTER TABLE job_audit ADD COLUMN request_hash TEXT;
ALTER TABLE job_audit ADD COLUMN request_sequence INTEGER NOT NULL DEFAULT 0;
ALTER TABLE job_audit ADD COLUMN request_json TEXT;
ALTER TABLE job_audit ADD COLUMN owner_id TEXT;
ALTER TABLE job_audit ADD COLUMN lease_until TEXT;
ALTER TABLE job_audit ADD COLUMN attempt INTEGER NOT NULL DEFAULT 0;
ALTER TABLE job_audit ADD COLUMN fencing_token INTEGER NOT NULL DEFAULT 0;
ALTER TABLE job_audit ADD COLUMN available_at TEXT NOT NULL DEFAULT '1970-01-01T00:00:00Z';
ALTER TABLE job_audit ADD COLUMN cancel_requested INTEGER NOT NULL DEFAULT 0;
INSERT OR IGNORE INTO tenant (id, name, slug) VALUES ('00000000-0000-0000-0000-000000000000', 'EvoBGP system jobs', 'evobgp-system-jobs');
CREATE INDEX idx_job_audit_claim ON job_audit (available_at, created_at) WHERE status = 'queued';
CREATE INDEX idx_job_audit_expired_lease ON job_audit (lease_until) WHERE status = 'running';
UPDATE job_audit SET owner_id = 'legacy-unfenced', lease_until = strftime('%Y-%m-%dT%H:%M:%fZ', 'now', '+15 minutes'), attempt = 1
WHERE status = 'running' AND lease_until IS NULL;
@@ -0,0 +1,7 @@
DROP TABLE deployment_speaker;
ALTER TABLE bgp_speaker DROP COLUMN deployment_job_id;
ALTER TABLE bgp_speaker DROP COLUMN deployment_request_sequence;
ALTER TABLE bgp_speaker DROP COLUMN deployment_generation;
ALTER TABLE bgp_speaker DROP COLUMN delivery_owner;
ALTER TABLE bgp_speaker DROP COLUMN delivery_token;
ALTER TABLE bgp_speaker DROP COLUMN delivery_until;
@@ -0,0 +1,27 @@
ALTER TABLE bgp_speaker ADD COLUMN deployment_generation INTEGER NOT NULL DEFAULT 0;
ALTER TABLE bgp_speaker ADD COLUMN deployment_request_sequence INTEGER NOT NULL DEFAULT 0;
ALTER TABLE bgp_speaker ADD COLUMN deployment_job_id TEXT;
ALTER TABLE bgp_speaker ADD COLUMN delivery_owner TEXT;
ALTER TABLE bgp_speaker ADD COLUMN delivery_token INTEGER NOT NULL DEFAULT 0;
ALTER TABLE bgp_speaker ADD COLUMN delivery_until TEXT;
CREATE TABLE deployment_speaker (
job_id TEXT NOT NULL REFERENCES job_audit(id) ON DELETE CASCADE,
tenant_id TEXT NOT NULL REFERENCES tenant(id) ON DELETE CASCADE,
speaker_id TEXT NOT NULL REFERENCES bgp_speaker(id) ON DELETE CASCADE,
revision_id TEXT NOT NULL,
generation INTEGER NOT NULL,
request_sequence INTEGER NOT NULL,
state TEXT NOT NULL,
confirmation TEXT NOT NULL DEFAULT '',
error_message TEXT NOT NULL DEFAULT '',
health_received_at TEXT,
health_observation_token INTEGER NOT NULL DEFAULT 0,
health_revision_id TEXT NOT NULL DEFAULT '',
health_source_at TEXT NOT NULL DEFAULT '',
health_error TEXT NOT NULL DEFAULT '',
retry_round INTEGER NOT NULL DEFAULT 0,
dispatch_attempts INTEGER NOT NULL DEFAULT 0,
updated_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY(job_id,speaker_id)
);
CREATE INDEX deployment_speaker_tenant_job ON deployment_speaker(tenant_id,job_id);
@@ -0,0 +1 @@
DROP TABLE telemetry_snapshot;
@@ -0,0 +1,21 @@
CREATE TABLE telemetry_snapshot (
tenant_id TEXT NOT NULL REFERENCES tenant(id) ON DELETE CASCADE,
target_id TEXT NOT NULL,
speaker_id TEXT REFERENCES bgp_speaker(id) ON DELETE SET NULL,
kind TEXT NOT NULL CHECK (kind IN ('remote_agent','local_bird')),
local_authority TEXT NOT NULL DEFAULT '',
config_fingerprint TEXT NOT NULL DEFAULT '',
bgp_json TEXT NOT NULL DEFAULT '{}',
health_json TEXT NOT NULL DEFAULT '{}',
next_poll_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP,
failure_count INTEGER NOT NULL DEFAULT 0,
discovery_cursor INTEGER NOT NULL DEFAULT 0,
last_refresh_requested_at TEXT,
refresh_pending INTEGER NOT NULL DEFAULT 0,
owner_id TEXT,
fencing_token INTEGER NOT NULL DEFAULT 0,
lease_until TEXT,
PRIMARY KEY (tenant_id,target_id)
);
CREATE INDEX telemetry_snapshot_due ON telemetry_snapshot(next_poll_at) WHERE owner_id IS NULL;
CREATE INDEX telemetry_snapshot_speaker ON telemetry_snapshot(speaker_id);
@@ -0,0 +1,3 @@
DROP TABLE IF EXISTS render_revision_manifest;
ALTER TABLE module_prefix_snapshot DROP COLUMN snapshot_version;
DROP TABLE IF EXISTS tenant_render_generation;
@@ -0,0 +1,14 @@
-- SQLite is used for schema parity and local fixtures; the production render
-- admission barrier is implemented by the PostgreSQL repository.
CREATE TABLE tenant_render_generation (
tenant_id TEXT PRIMARY KEY REFERENCES tenant(id) ON DELETE CASCADE,
generation INTEGER NOT NULL DEFAULT 1
);
INSERT INTO tenant_render_generation(tenant_id, generation) SELECT id, 1 FROM tenant;
ALTER TABLE module_prefix_snapshot ADD COLUMN snapshot_version INTEGER NOT NULL DEFAULT 0;
CREATE TABLE render_revision_manifest (
revision_id TEXT PRIMARY KEY REFERENCES config_revision(id) ON DELETE CASCADE,
tenant_id TEXT NOT NULL REFERENCES tenant(id) ON DELETE CASCADE,
config_generation INTEGER NOT NULL,
manifest_json TEXT NOT NULL
);