Add support for host-specific VSPC policies, and HTTP retries to increase reliability.

vspc_backup_checks/2.3/local/lib/python3/cmk_addons/plugins/vspc_backup_checks/libexec/agent_vspc_backup_checks

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 # Copyright (C) 2026 Spearhead Systems SRL
 
-import http.client, argparse, json, ssl
+import http.client, argparse, json, ssl, time
 from datetime import datetime, timezone
 from collections import defaultdict
 
@@ -16,6 +16,26 @@ DAILY  = "Daily"
 WEEKLY = "Weekly"
 
 
+# Attempt to GET a URL, retrying with exponential back-off every time a 429
+# is returned. 429 is what VSPC returns when its rate limiting kicks in.
+def get_url(conn, path, headers):
+    delay = 2   # seconds
+
+    while delay <= 64:
+        conn.request("GET", path, headers=headers)
+        response = conn.getresponse()
+
+        if response.status != 429:
+            return response
+
+        response.read()
+
+        time.sleep(delay)
+        delay *= 2
+
+    return
+
+
 # GET HTTP with Bearer auth. Returns data structure parsed from JSON.
 #
 # Since results in VSPC are paginated, we go through all pages and return an
@@ -36,17 +56,20 @@ def get_paginated_json_url(host, port, path, token, insecure):
     offset = 0
 
     while True:
-        conn.request("GET", f"{path}?offset={offset}", headers=headers)
-        response = conn.getresponse()
+        response = get_url(conn, f"{path}?offset={offset}", headers)
 
-        if response.status != 200:
+        if not response or response.status != 200:
             raise Exception(f"Status code for {path} was {response.status}")
 
         page = json.loads(response.read())
 
-        meta = page["meta"]
+        meta = page.get("meta")
         data = page["data"]
 
+        # If not meta, we're not paginated, so return the data immediately
+        if not meta:
+            return data
+
         results.extend(data)
 
         total  = meta["pagingInfo"]["total"]
@@ -124,7 +147,7 @@ def parse_arguments():
 #     ],
 #     ...
 # }
-def process(mAgents, bAgents, jobs, restores, policies):
+def process(mAgents, bAgents, jobs, restores, policies, get_customized_policy):
     mToB = {}
     for agent in bAgents:
         mToB[agent["managementAgentUid"]] = agent
@@ -228,6 +251,22 @@ def process(mAgents, bAgents, jobs, restores, policies):
             critDays = 0
             if sched == DAILY:
                 policyType = policies.get(polId)
+
+                # There's no policyType when a customized policy is used,
+                # since the policy is specific to a single backup agent and job.
+                # That means we must fetch this customized policy here.
+                if policyType is None:
+                    customizedPolicy = get_customized_policy(bAgent, jobId)
+                    jobConfig = customizedPolicy["jobConfiguration"]
+                    customizedSched = jobConfig.get("scheduleSettings")
+
+                    if jobConfig.get("serverModeSettings"):
+                        customizedSched = jobConfig["serverModeSettings"]["scheduleSetting"]
+                    elif jobConfig.get("workstationModeSettings"):
+                        customizedSched = jobConfig["workstationModeSettings"]["scheduleSetting"]
+
+                    policyType = get_period(customizedSched)
+
                 if policyType == WEEKLY:
                     # It may seem silly to check for WEEKLY under a DAILY section,
                     # but the scheduling allows for this, and we actually use this
@@ -310,13 +349,7 @@ def process(mAgents, bAgents, jobs, restores, policies):
 # run certain backup jobs only on a Saturday.
 #
 # So we need to filter through VSPC's somewhat inconsistent policy API to find
-# daily settings. If a job is supposed to run on exactly one day a week, we
-# treat it as weekly, otherwise daily.
-#
-# NB: if we ever use more than one day during the week (e.g. two), the logic in
-# this plugin will need to be changed. The logic here, and in process() above,
-# match what we currently do in prod, since a "better" approach will require a
-# lot more code for a system we might phase out for something else.
+# daily settings.
 def mergePolicies(linuxPolicies, windowsPolicies, macPolicies):
     policies = {}
     schedules = {}
@@ -339,19 +372,32 @@ def mergePolicies(linuxPolicies, windowsPolicies, macPolicies):
         schedules[policy["instanceUid"]] = schedule
 
     for polId, sched in schedules.items():
-        dailySched = sched.get("dailyScheduleSettings")
-        if not dailySched:
-            continue
-
-        days = dailySched.get("specificDays")
-        if days and len(days) == 1:
-            policies[polId] = WEEKLY
-        else:
-            policies[polId] = DAILY
+        period = get_period(sched)
+        if period:
+            policies[polId] = period
 
     return policies
 
 
+# Return whether a policy schedule is DAILY or WEEKLY. If a job is supposed to
+# run on exactly one day a week, we treat it as weekly, otherwise daily.
+#
+# NB: if we ever use more than one day during the week (e.g. two), the logic in
+# this plugin will need to be changed. The logic here, and in process() above,
+# match what we currently do in prod, since a "better" approach will require a
+# lot more code for a system we might phase out for something else.
+def get_period(sched):
+    dailySched = sched.get("dailyScheduleSettings")
+    if not dailySched:
+        return
+
+    days = dailySched.get("specificDays")
+    if days and len(days) == 1:
+        return WEEKLY
+    else:
+        return DAILY
+
+
 def print_demo():
     print("""
 <<<<newveeam>>>>
@@ -401,6 +447,11 @@ def main(argv=None):
     if args.demo:
         return print_demo()
 
+    def get_customized_policy(bAgent, jobId):
+        os  = bAgent["agentPlatform"].lower()
+        bId = bAgent["instanceUid"]
+        return get_paginated_json_url(args.hostname, args.port, f"/api/v3/infrastructure/backupAgents/{os}/{bId}/jobs/{jobId}/configuration", args.token, args.insecure)
+
     linuxPolicies   = get_paginated_json_url(args.hostname, args.port, '/api/v3/configuration/backupPolicies/linux',   args.token, args.insecure)
     windowsPolicies = get_paginated_json_url(args.hostname, args.port, '/api/v3/configuration/backupPolicies/windows', args.token, args.insecure)
     macPolicies     = get_paginated_json_url(args.hostname, args.port, '/api/v3/configuration/backupPolicies/mac',     args.token, args.insecure)
@@ -412,7 +463,8 @@ def main(argv=None):
 
     policies = mergePolicies(linuxPolicies, windowsPolicies, macPolicies)
 
-    results = process(mAgents, bAgents, jobs, restores, policies)
+
+    results = process(mAgents, bAgents, jobs, restores, policies, get_customized_policy)
     print_out(results)