conductor(checkpoint): Checkpoint end of Phase 2

This commit is contained in:
2026-02-08 07:54:49 -08:00
parent a686c5b225
commit 6d77729a4a
7 changed files with 196 additions and 6 deletions

View File

@@ -1,9 +1,10 @@
import consul_client
import litefs_client
import nomad_client
def get_cluster_status(consul_url):
def get_cluster_status(consul_url, job_id="navidrome-litefs"):
"""
Aggregates cluster data from Consul and LiteFS.
Aggregates cluster data from Consul, LiteFS, and Nomad.
"""
consul_nodes = consul_client.get_cluster_services(consul_url)
aggregated_nodes = []
@@ -21,11 +22,17 @@ def get_cluster_status(consul_url):
"uptime": litefs_status.get("uptime", "N/A"),
"advertise_url": litefs_status.get("advertise_url", ""),
"replication_lag": litefs_status.get("replication_lag", "N/A"),
"litefs_error": litefs_status.get("error", None)
"litefs_error": litefs_status.get("error", None),
"nomad_logs": None
}
if node["status"] != "passing":
is_healthy = False
# Fetch Nomad logs for critical nodes
alloc_id = nomad_client.get_allocation_id(node["node"], job_id)
if alloc_id:
node_data["alloc_id"] = alloc_id
node_data["nomad_logs"] = nomad_client.get_allocation_logs(alloc_id)
if node_data["litefs_primary"]:
primary_count += 1