remove unique version check in monrun ch_replication_lag and ch_system_queues

ch_tools/monrun_checks/ch_replication_lag.py

@@ -1,12 +1,10 @@
-import logging
 from typing import Any, Dict
 
 import click
 from tabulate import tabulate
 
 from ch_tools.common.result import Result
 from ch_tools.monrun_checks.clickhouse_client import ClickhouseClient
-from ch_tools.monrun_checks.clickhouse_info import ClickhouseInfo
 
 
 @click.command("replication-lag")
@@ -160,16 +158,6 @@ def replication_lag_command(ctx, xcrit, crit, warn, mwarn, mcrit, verbose):
         lag, lag_with_errors, max_execution, max_merges
     )
 
-    try:
-        replica_versions_mismatch = ClickhouseInfo.get_versions_count(ctx) > 1
-        if replica_versions_mismatch:
-            msg += ", ClickHouse versions on replicas mismatch"
-            return Result(code=1, message=msg, verbose=msg_verbose)
-    except Exception:
-        logging.warning("Unable to get version info from replicas", exc_info=True)
-        msg += ", one or more replicas is unavailable"
-        return Result(code=1, message=msg, verbose=msg_verbose)
-
     if (
         lag_with_errors < crit
         and max_execution < xcrit

ch_tools/monrun_checks/ch_system_queues.py

@@ -3,7 +3,6 @@
 
 from ch_tools.common.result import Result
 from ch_tools.monrun_checks.clickhouse_client import ClickhouseClient
-from ch_tools.monrun_checks.clickhouse_info import ClickhouseInfo
 
 
 @click.command("system-queues")
@@ -27,7 +26,7 @@ def system_queues_command(ctx, crit, warn, conf):
         config = {"triggers": {"default": {"crit": crit, "warn": warn}}}
 
     metrics = get_metrics(ctx)
-    return check_metrics(ctx, metrics, config)
+    return check_metrics(metrics, config)
 
 
 def get_config(conf):
@@ -49,7 +48,7 @@ def get_metrics(ctx):
     return ClickhouseClient(ctx).execute(query, compact=False)
 
 
-def check_metrics(ctx, metrics, config):
+def check_metrics(metrics, config):
     """
     Check that metrics are within acceptable levels.
     """
@@ -60,7 +59,6 @@ def check_metrics(ctx, metrics, config):
     status = 0
     message = "OK"
     triggers = []
-    versions_count = 0
 
     for row in metrics:
         db_table = "{}.{}".format(row["database"], row["table"])
@@ -73,17 +71,10 @@ def check_metrics(ctx, metrics, config):
             thresholds = table_thresholds.get(key, default_thresholds[key])
 
             report = ""
-            table_status = 0
             for prior in "crit", "warn":
                 threshold = thresholds[prior]
                 if value > threshold:
                     table_status = status_map[prior]
-                    if table_status > 1:
-                        if versions_count == 0:
-                            versions_count = ClickhouseInfo.get_versions_count(ctx)
-                        if versions_count > 1:
-                            table_status = 1
-                            prior = "crit->warn"
                     report += "{}: {} {} > {} ({});".format(
                         db_table, key, value, threshold, prior
                     )
@@ -94,9 +85,5 @@ def check_metrics(ctx, metrics, config):
         triggers.sort(reverse=True, key=lambda x: x[0])
         status = triggers[0][0]
         message = " ".join(x[1] for x in triggers)
-        if versions_count == 0:
-            versions_count = ClickhouseInfo.get_versions_count(ctx)
-        if versions_count > 1:
-            message += " ClickHouse versions on replicas mismatch"
 
     return Result(code=status, message=message)

ch_tools/monrun_checks/clickhouse_info.py

@@ -1,26 +1,9 @@
 import functools
 
-from ch_tools.monrun_checks.clickhouse_client import ClickhouseClient, ClickhousePort
+from ch_tools.monrun_checks.clickhouse_client import ClickhouseClient
 
 
 class ClickhouseInfo:
-    @classmethod
-    @functools.lru_cache(maxsize=1)
-    def get_versions_count(cls, ctx):
-        """
-        Count different clickhouse versions in cluster.
-        """
-        ch_client = ClickhouseClient(ctx)
-        #  I believe that all hosts in cluster have the same port set, so check current for security port
-        remote_command = (
-            "remoteSecure"
-            if ch_client.check_port(ClickhousePort.TCP_SECURE)
-            else "remote"
-        )
-        replicas = ",".join(cls.get_replicas(ctx))
-        query = f"""SELECT uniq(version()) FROM {remote_command}('{replicas}', system.one)"""
-        return int(ch_client.execute(query)[0][0])
-
     @classmethod
     @functools.lru_cache(maxsize=1)
     def get_replicas(cls, ctx):