By hashing the config values and files, only restart services on differences

charms/kubernetes_snaps.py

@@ -1,3 +1,5 @@
+import contextlib
+import hashlib
 import ipaddress
 import json
 import logging
@@ -100,7 +102,6 @@ def configure_apiserver(
     enc_provider_config = encryption_config_path()
     if enc_provider_config.exists():
         api_opts["encryption-provider-config"] = str(enc_provider_config)
-
     api_opts["advertise-address"] = advertise_address
 
     api_opts["etcd-cafile"] = "/root/cdk/etcd/client-ca.pem"
@@ -192,7 +193,36 @@ def configure_apiserver(
     else:
         remove_if_exists(audit_webhook_conf_path)
 
-    configure_kubernetes_service("kube-apiserver", api_opts, extra_args_config)
+    config_files = set(
+        map(
+            api_opts.get,
+            (
+                "tls-cert-file",
+                "tls-private-key-file",
+                "kubelet-certificate-authority",
+                "kubelet-client-certificate",
+                "kubelet-client-key",
+                "authentication-token-webhook-config-file",
+                "service-account-signing-key-file",
+                "service-account-key-file",
+                "encryption-provider-config",
+                "etcd-cafile",
+                "etcd-keyfile",
+                "etcd-certfile",
+                "authorization-webhook-config-file",
+                "requestheader-client-ca-file",
+                "proxy-client-cert-file",
+                "proxy-client-key-file",
+                "client-ca-file",
+                "audit-policy-file",
+                "audit-webhook-config-file",
+            ),
+        )
+    )
+
+    configure_kubernetes_service(
+        "kube-apiserver", api_opts, extra_args_config, config_files
+    )
 
 
 def configure_controller_manager(
@@ -231,10 +261,26 @@ def configure_controller_manager(
 
     controller_opts["feature-gates"] = ",".join(feature_gates)
 
+    config_files = set(
+        map(
+            controller_opts.get,
+            (
+                "root-ca-file",
+                "kubeconfig",
+                "authorization-kubeconfig",
+                "authentication-kubeconfig",
+                "service-account-private-key-file",
+                "tls-cert-file",
+                "tls-private-key-file",
+            ),
+        )
+    )
+
     configure_kubernetes_service(
         "kube-controller-manager",
         controller_opts,
         extra_args_config,
+        config_files,
     )
 
 
@@ -286,10 +332,13 @@ def configure_kube_proxy(
         f.write("# Generated by kubernetes-common library, do not edit\n")
         yaml.dump(kube_proxy_config, f)
 
+    config_files = {
+        kube_proxy_opts["config"],
+        kubeconfig,
+    }
+
     configure_kubernetes_service(
-        "kube-proxy",
-        kube_proxy_opts,
-        extra_args_config,
+        "kube-proxy", kube_proxy_opts, extra_args_config, config_files
     )
 
 
@@ -375,38 +424,23 @@ def configure_kubelet(
         f.write("# Generated by charm, do not edit\n")
         yaml.dump(kubelet_config, f)
 
+    config_files = {
+        kubelet_opts["config"],
+        kubeconfig,
+        "/run/systemd/resolve/resolv.conf",
+        "/root/cdk/ca.crt",
+        "/root/cdk/server.crt",
+        "/root/cdk/server.key",
+    }
+
     configure_kubernetes_service(
         "kubelet",
         kubelet_opts,
         extra_args_config,
+        config_files,
     )
 
 
-def configure_kubernetes_service(service, base_args, extra_args_config):
-    extra_args = parse_extra_args(extra_args_config)
-
-    args = {}
-    args.update(base_args)
-    args.update(extra_args)
-
-    # TODO: CIS arg handling???
-    # CIS benchmark action may inject kv config to pass failing tests. Merge
-    # these after the func args as they should take precedence.
-    # cis_args_key = "cis-" + service
-    # cis_args = db.get(cis_args_key) or {}
-    # args.update(cis_args)
-
-    # Remove any args with 'None' values (all k8s args are 'k=v') and
-    # construct an arg string for use by 'snap set'.
-    args = {k: v for k, v in args.items() if v is not None}
-    args = ['--%s="%s"' % arg for arg in args.items()]
-    args = " ".join(args)
-
-    cmd = ["snap", "set", service, f"args={args}"]
-    check_call(cmd)
-    service_restart(f"snap.{service}.daemon")
-
-
 def configure_scheduler(extra_args_config, kubeconfig):
     kube_scheduler_config_path = "/root/cdk/kube-scheduler-config.yaml"
     scheduler_opts = {}
@@ -433,7 +467,106 @@ def configure_scheduler(extra_args_config, kubeconfig):
     with open(kube_scheduler_config_path, "w") as f:
         yaml.safe_dump(scheduler_config, f)
 
-    configure_kubernetes_service("kube-scheduler", scheduler_opts, extra_args_config)
+    config_files = {
+        scheduler_opts["config"],
+        kubeconfig,
+        kube_scheduler_config_path,
+    }
+
+    configure_kubernetes_service(
+        "kube-scheduler", scheduler_opts, extra_args_config, config_files
+    )
+
+
+def _sha256_file(config_file: str) -> hashlib.sha256:
+    h = hashlib.sha256()
+    h.update(config_file.encode())
+    config_file = Path(config_file)
+    if Path.is_file(config_file):
+        with config_file.open("rb") as file:
+            while True:
+                # Reading is buffered, so we can read smaller chunks.
+                chunk = file.read(h.block_size)
+                if not chunk:
+                    break
+                h.update(chunk)
+    else:
+        h.update(b"--empty--")
+    return h
+
+
+def _dict_compare(d1, d2):
+    d1_keys = set(d1.keys())
+    d2_keys = set(d2.keys())
+    shared_keys = d1_keys.intersection(d2_keys)
+    added = d1_keys - d2_keys
+    removed = d2_keys - d1_keys
+    modified = {o: (d1[o], d2[o]) for o in shared_keys if d1[o] != d2[o]}
+    same = set(o for o in shared_keys if d1[o] == d2[o])
+    return added, removed, modified, same
+
+
+@contextlib.contextmanager
+def _calculate_config_difference(service: str, args, config_files):
+    args_hash = _snap_common_path(service) / "args-hash.txt"
+
+    # gather existing config hash
+    current, updated = {}, {}
+    if Path.is_file(args_hash):
+        try:
+            current.update(yaml.safe_load(args_hash.open()))
+        except yaml.YAMLError:
+            log.warning("Failed to load existing args hash for %s", service)
+
+    # calculate new config hash
+    updated = {arg: hashlib.sha256(str(val).encode()) for arg, val in args.items()}
+    updated.update({f"path:{key}": _sha256_file(key) for key in config_files if key})
+    updated = {k: v.hexdigest() for k, v in updated.items()}
+
+    # discern differences
+    added, removed, modified, _ = _dict_compare(updated, current)
+    for key in added:
+        log.debug("%s:   Added config value %s", service.capitalize(), key)
+    for key in removed:
+        log.debug("%s: Dropped config value %s", service.capitalize(), key)
+    for key in modified:
+        log.debug("%s: Updated config value %s", service.capitalize(), key)
+
+    different = added or removed or modified
+    yield different
+    if different:
+        yaml.safe_dump(updated, args_hash.open("w"))
+    else:
+        log.debug("%s: No config changes detected", service.capitalize())
+
+
+def configure_kubernetes_service(
+    service, base_args, extra_args_config, config_files=None
+):
+    extra_args = parse_extra_args(extra_args_config)
+
+    args = {}
+    args.update(base_args)
+    args.update(extra_args)
+
+    # TODO: CIS arg handling???
+    # CIS benchmark action may inject kv config to pass failing tests. Merge
+    # these after the func args as they should take precedence.
+    # cis_args_key = "cis-" + service
+    # cis_args = db.get(cis_args_key) or {}
+    # args.update(cis_args)
+
+    # Remove any args with 'None' values (all k8s args are 'k=v') and
+    # construct an arg string for use by 'snap set'.
+
+    args = {k: v for k, v in args.items() if v is not None}
+    with _calculate_config_difference(service, args, config_files or {}) as restart:
+        if restart:
+            args = ['--%s="%s"' % arg for arg in args.items()]
+            args = " ".join(args)
+            cmd = ["snap", "set", service, f"args={args}"]
+            check_call(cmd)
+            service_restart(f"snap.{service}.daemon")
 
 
 def configure_services_restart_always(control_plane=False):