From: Vasek Sraier <git@vakabus.cz>
Date: Fri, 5 Aug 2022 13:43:38 +0000 (+0200)
Subject: manager: improve handling of max-workers configuration
X-Git-Tag: v6.0.0a1~27^2~1
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=51ee4f7bea41e9895f4cf2b92d4f02453ee77f36;p=thirdparty%2Fknot-resolver.git

manager: improve handling of max-workers configuration

- changed default to 10*#cpus or 256 if not possible
- fixed bug that would prevent us from running the limiting number of workers

closes #753
---

diff --git a/manager/knot_resolver_manager/datamodel/config_schema.py b/manager/knot_resolver_manager/datamodel/config_schema.py
index 26194b049..4941d6ade 100644
--- a/manager/knot_resolver_manager/datamodel/config_schema.py
+++ b/manager/knot_resolver_manager/datamodel/config_schema.py
@@ -71,6 +71,13 @@ def _cpu_count() -> Optional[int]:
         return cpus
 
 
+def _default_max_worker_count() -> Optional[int]:
+    c = _cpu_count()
+    if c is not None:
+        return c * 10
+    return MAX_WORKERS
+
+
 class KresConfig(BaseSchema):
     class Raw(BaseSchema):
         """
@@ -105,7 +112,7 @@ class KresConfig(BaseSchema):
         hostname: Optional[str] = None
         rundir: UncheckedPath = UncheckedPath(".")
         workers: Union[Literal["auto"], IntPositive] = IntPositive(1)
-        max_workers: IntPositive = IntPositive(MAX_WORKERS)
+        max_workers: IntPositive = IntPositive(_default_max_worker_count())
         management: ManagementSchema = ManagementSchema({"unix-socket": "./manager.sock"})
         webmgmt: Optional[WebmgmtSchema] = None
         options: OptionsSchema = OptionsSchema()
@@ -176,12 +183,16 @@ class KresConfig(BaseSchema):
         return obj.dns64
 
     def _validate(self) -> None:
-        cpu_count = _cpu_count()
+        # enforce max-workers config
+        if int(self.workers) > int(self.max_workers):
+            raise ValueError(f"can't run with more workers then the configured maximum {self.max_workers}")
 
+        # sanity check
+        cpu_count = _cpu_count()
         if cpu_count and int(self.workers) > 10 * cpu_count:
-            raise ValueError("refusing to run with more then 10 workers per cpu core")
-        elif int(self.workers) > MAX_WORKERS:
-            raise ValueError(f"refusing to run with more workers then allowed maximum {MAX_WORKERS}")
+            raise ValueError(
+                "refusing to run with more then 10 workers per cpu core, the system wouldn't behave nicely"
+            )
 
     def render_lua(self) -> str:
         # FIXME the `cwd` argument is used only for configuring control socket path
diff --git a/manager/knot_resolver_manager/kresd_controller/supervisord/config_file.py b/manager/knot_resolver_manager/kresd_controller/supervisord/config_file.py
index bbce2bc11..1fd1e1b56 100644
--- a/manager/knot_resolver_manager/kresd_controller/supervisord/config_file.py
+++ b/manager/knot_resolver_manager/kresd_controller/supervisord/config_file.py
@@ -73,7 +73,7 @@ class ProcessTypeConfig:
             workdir=cwd,
             command=f"{kresd_executable()} -c {kresd_config_file_supervisord_pattern(config)} -n",
             environment='SYSTEMD_INSTANCE="%(process_num)d",X-SUPERVISORD-TYPE=notify',
-            max_procs=config.max_workers,
+            max_procs=int(config.max_workers) + 1,  # +1 for the canary process
         )
 
     @staticmethod