improvement(gemini): run gemini from scylladb/gemini

Dockerfile has been moved to gemini project, and now SCT should use that image. Changes: - Use image from scylladb/gemini - Remove dockerfile for gemini and point in readme location for the images - Add CQL Statement Logging to gemini output - Forward outputs from docker to $HOME/*.log - Run default gemini flags from gemini_thread.py Signed-off-by: Dusan Malusev <[email protected]>
scylladb · Jan 10, 2025 · 0822faa · 0822faa
1 parent 6355082
commit 0822faa
Show file tree

Hide file tree

Showing 20 changed files with 189 additions and 169 deletions.
diff --git a/defaults/docker_images/gemini/values_gemini.yaml b/defaults/docker_images/gemini/values_gemini.yaml
@@ -1,2 +1,2 @@
 gemini:
-  image: scylladb/hydra-loaders:gemini-v1.8.6
+  image: scylladb/gemini:1.8.9
diff --git a/defaults/test_default.yaml b/defaults/test_default.yaml
@@ -90,7 +90,6 @@ alternator_access_key_id: ''
 alternator_secret_access_key: ''
 dynamodb_primarykey_type: 'HASH'
 
-
 store_cdclog_reader_stats_in_es: false
 region_aware_loader: false
 
@@ -158,7 +157,7 @@ jepsen_test_cmd:
 jepsen_test_count: 1
 jepsen_test_run_policy: all
 
-max_events_severities: ""
+max_events_severities: ''
 scylla_mgmt_agent_version: '3.4.0'
 mgmt_docker_image: 'scylladb/scylla-manager:3.4.0'
 k8s_log_api_calls: false
@@ -283,7 +282,7 @@ latency_decorator_error_thresholds:
   mixed:
     default:
       P90 write:
-        fixed_limit: 5
+          fixed_limit: 5
       P90 read:
           fixed_limit: 5
       P99 write:

diff --git a/docker/gemini/Dockerfile b/docker/gemini/Dockerfile
diff --git a/docker/gemini/README.md b/docker/gemini/README.md
@@ -1,11 +1,6 @@
-Currently, when releasing a new version of Gemini, there's no need to push the image to Docker Hub.
-The image is built and pushed automatically by `goreleaser` when a new version is released.
-Docs from Gemini repo: https://github.com/scylladb/gemini/blob/master/docs/release-process.md
-Steps to release gemini :
-```
-0. Make sure you have proper go installed. See the version in https://github.com/scylladb/gemini/blob/master/go.mod
-1. update changelog and tag the commit
-2. create github token with write:packages permissions here: https://github.com/settings/tokens/new
-3. export GITHUB_TOKEN="YOUR_GH_TOKEN”
-4. Run `goreleaser`from`cmd/gemini`directory
-```
+# Gemini Image
+
+## Locations
+
+- [DockerHub](https://hub.docker.com/r/scylladb/gemini)
+- [Gemini Github](https://github.com/scylladb/gemini)
diff --git a/docker/gemini/image b/docker/gemini/image
diff --git a/sdcm/gemini_thread.py b/sdcm/gemini_thread.py
@@ -66,50 +66,105 @@ def __init__(self, test_cluster, oracle_cluster, loaders, stress_cmd, timeout=No
         super().__init__(loader_set=loaders, stress_cmd=stress_cmd, timeout=timeout, params=params)
         self.test_cluster = test_cluster
         self.oracle_cluster = oracle_cluster
-        self._gemini_result_file = None
         self.gemini_commands = []
-        self.gemini_request_timeout = 180
-        self.gemini_connect_timeout = 120
-
-    @property
-    def gemini_result_file(self):
-        if not self._gemini_result_file:
-            self._gemini_result_file = os.path.join("/", "gemini_result_{}.log".format(uuid.uuid4()))
-        return self._gemini_result_file
+        self.unique_id = uuid.uuid4()
+        self.gemini_default_flags = {
+            "level": "info",
+            "request-timeout": "60s",
+            "connect-timeout": "60s",
+            "consistency": "QUORUM",
+            "async-objects-stabilization-backoff": "1s",
+            "async-objects-stabilization-attempts": 10,
+            "max-mutation-retries-backoff": "1s",
+            "max-mutation-retries": 10,
+            "dataset-size": "large",
+            "oracle-host-selection-policy": "token-aware",
+            "test-host-selection-policy": "token-aware",
+            "drop-schema": "true",
+            "cql-features": "normal",
+            "fail-fast": "true",
+            "materialized-views": "false",
+            "use-server-timestamps": "true",
+            "use-lwt": "false",
+            "use-counters": "false",
+            "max-tables": 1,
+            "max-columns": 16,
+            "min-columns": 8,
+            "max-partition-keys": 6,
+            "min-partition-keys": 2,
+            "max-clustering-keys": 4,
+            "min-clustering-keys": 2,
+            "partition-key-distribution": "normal",  # Distribution for hitting the partition
+            # These two are used to control the memory usage of Gemini
+            "token-range-slices": 512,  # Number of partitions
+            "partition-key-buffer-reuse-size": 100,  # Internal Channel Size per parittion value generation
+        }
+
+        self.gemini_oracle_statements_file = f"gemini_oracle_statements_{self.unique_id}.log"
+        self.gemini_test_statements_file = f"gemini_test_statements_{self.unique_id}.log"
+        self.gemini_result_file = f"gemini_result_{self.unique_id}.log"
 
     def _generate_gemini_command(self):
-        seed = self.params.get('gemini_seed')
+        seed = self.params.get('gemini_seed') or random.randint(1, 100)
         table_options = self.params.get('gemini_table_options')
-        if not seed:
-            seed = random.randint(1, 100)
+        log_statements = self.params.get('gemini_log_cql_statements') or False
+
         test_nodes = ",".join(self.test_cluster.get_node_cql_ips())
-        oracle_nodes = ",".join(self.oracle_cluster.get_node_cql_ips()) if self.oracle_cluster else None
-
-        cmd = "./{} --test-cluster={} --outfile {} --seed {} --request-timeout {}s --connect-timeout {}s ".format(
-            self.stress_cmd.strip(),
-            test_nodes,
-            self.gemini_result_file,
-            seed,
-            self.gemini_request_timeout,
-            self.gemini_connect_timeout)
-        if oracle_nodes:
-            cmd += "--oracle-cluster={} ".format(oracle_nodes)
+        oracle_nodes = ",".join(self.oracle_cluster.get_node_cql_ips())
+
+        cmd = f"gemini \
+                --non-interactive \
+                --oracle-cluster=\"{oracle_nodes}\" \
+                --test-cluster=\"{test_nodes}\" \
+                --seed={seed} \
+                --schema-seed={seed} \
+                --profiling-port=6060 \
+                --bind=0.0.0.0:2121 \
+                --outfile=/{self.gemini_result_file} \
+                --replication-strategy=\"{{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}}\" \
+                --oracle-replication-strategy=\"{{'class': 'NetworkTopologyStrategy', 'replication_factor': '1'}}\" "
+
+        if log_statements:
+            cmd += f"--test-statement-log-file=/{self.gemini_test_statements_file} \
+                    --oracle-statement-log-file=/{self.gemini_oracle_statements_file} "
+
+        credentials = self.loader_set.get_db_auth()
+
+        if credentials and '--test-username' not in cmd:
+            cmd += f"--test-username={credentials[0]} \
+                --test-password={credentials[1]} \
+                --oracle-username={credentials[0]} \
+                --oracle-password={credentials[1]} "
+
         if table_options:
-            cmd += " ".join([f"--table-options \"{table_opt}\"" for table_opt in table_options])
+            cmd += " ".join([f"--table-options=\"{table_opt}\"" for table_opt in table_options])
+
+        stress_cmd = self.stress_cmd.replace('\n', ' ').strip()
+
+        for key, value in self.gemini_default_flags.items():
+            if not key in stress_cmd:
+                cmd += f"--{key}={value} "
+
+        cmd += stress_cmd
         self.gemini_commands.append(cmd)
         return cmd
 
     def _run_stress(self, loader, loader_idx, cpu_idx):
-
-        cpu_options = ""
-        if self.stress_num > 1:
-            cpu_options = f'--cpuset-cpus="{cpu_idx}"'
-
-        docker = cleanup_context = RemoteDocker(loader, self.docker_image_name,
-                                                extra_docker_opts=f'{cpu_options} --label shell_marker={self.shell_marker} '
-                                                                  '--network=host '
-                                                                  '--security-opt seccomp=unconfined '
-                                                                  '--entrypoint=""')
+        for file_name in [self.gemini_result_file, self.gemini_test_statements_file, self.gemini_oracle_statements_file]:
+            loader.remoter.run(f"touch $HOME/{file_name}", ignore_status=True, verbose=False)
+
+        docker = cleanup_context = RemoteDocker(
+            loader,
+            self.docker_image_name,
+            extra_docker_opts=f'--cpuset-cpus="{cpu_idx}"' if self.stress_num > 1 else ""
+            '--label shell_marker={self.shell_marker}'
+            '--network=host '
+            '--security-opt seccomp=unconfined '
+            '--entrypoint="" '
+            f'-v $HOME/{self.gemini_result_file}:/{self.gemini_result_file} '
+            f'-v $HOME/{self.gemini_test_statements_file}:/{self.gemini_test_statements_file} '
+            f'-v $HOME/{self.gemini_oracle_statements_file}:/{self.gemini_oracle_statements_file} '
+        )
 
         if not os.path.exists(loader.logdir):
             os.makedirs(loader.logdir, exist_ok=True)
@@ -148,6 +203,13 @@ def _run_stress(self, loader, loader_idx, cpu_idx):
             results_copied = docker.receive_files(src=self.gemini_result_file, dst=local_gemini_result_file)
             assert results_copied, "gemini results aren't available, did gemini even run ?"
 
+            local_gemini_test_statements_file = os.path.join(
+                docker.node.logdir, os.path.basename(self.gemini_test_statements_file))
+            local_gemini_oracle_statements_file = os.path.join(
+                docker.node.logdir, os.path.basename(self.gemini_oracle_statements_file))
+            docker.receive_files(src=self.gemini_test_statements_file, dst=local_gemini_test_statements_file)
+            docker.receive_files(src=self.gemini_oracle_statements_file, dst=local_gemini_oracle_statements_file)
+
         return docker, result, local_gemini_result_file
 
     def get_gemini_results(self):

diff --git a/sdcm/logcollector.py b/sdcm/logcollector.py
@@ -847,6 +847,8 @@ class LoaderLogCollector(LogCollector):
                 search_locally=True),
         FileLog(name='*gemini-l*.log',
                 search_locally=True),
+        FileLog(name='*gemini_*_statements_*.log',
+                search_locally=True),
         FileLog(name='gemini_result*.log',
                 search_locally=True),
         FileLog(name='cdclogreader*.log',

diff --git a/sdcm/sct_config.py b/sdcm/sct_config.py
@@ -708,6 +708,9 @@ class SCTConfiguration(dict):
 
         dict(name="gemini_seed", env="SCT_GEMINI_SEED", type=int,
              help="Seed number for gemini command"),
+        dict(name="gemini_log_cql_statements",
+             env="SCT_GEMINI_LOG_CQL_STATEMENTS",
+             type=boolean, help="Log CQL statements to file"),
         dict(name="gemini_table_options", env="SCT_GEMINI_TABLE_OPTIONS", type=list,
              help="""table options for created table. example:
                      ["cdc={'enabled': true}"]

diff --git a/test-cases/gemini/gemini-1tb-10h.yaml b/test-cases/gemini/gemini-1tb-10h.yaml
@@ -13,15 +13,14 @@ nemesis_selector: ['run_with_gemini']
 nemesis_interval: 5
 nemesis_seed: '041'
 
-gemini_cmd: "gemini -d --duration 8h --warmup 2h -c 50 \
--m mixed -f --non-interactive --cql-features normal \
---max-mutation-retries 10 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms \
---replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}\"
---oracle-replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '1'}\" "
-
+gemini_cmd: |
+  --duration 8h
+  --warmup 2h
+  --concurrency 50
+  --mode mixed
 
 gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
+gemini_log_cql_statements: false
 
 db_type: mixed_scylla
 instance_type_db_oracle: 'i4i.16xlarge'
diff --git a/test-cases/gemini/gemini-3h-cdc-postimage-write.yaml b/test-cases/gemini/gemini-3h-cdc-postimage-write.yaml
@@ -4,17 +4,16 @@ n_loaders: 1
 n_monitor_nodes: 1
 instance_type_db: 'i4i.4xlarge'
 
-user_prefix: "gemini-cdc-postimage-write"
-
-gemini_cmd: "gemini -d --duration 3h \
--c 30 -m write -f --non-interactive --cql-features normal \
---max-mutation-retries 5 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms \
---replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}\" "
+user_prefix: 'gemini-cdc-postimage-write'
 
+gemini_cmd: |
+  --duration 3h
+  --concurrency 30
+  --mode write
 
 gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
 gemini_table_options:
   - "cdc={'enabled': true, 'postimage': true}"
+gemini_log_cql_statements: false
 
 db_type: scylla
diff --git a/test-cases/gemini/gemini-3h-cdc-preimage-write.yaml b/test-cases/gemini/gemini-3h-cdc-preimage-write.yaml
@@ -4,17 +4,16 @@ n_loaders: 1
 n_monitor_nodes: 1
 instance_type_db: 'i4i.4xlarge'
 
-user_prefix: "gemini-cdc-preimage-write"
-
-gemini_cmd: "gemini -d --duration 3h \
--c 30 -m write -f --non-interactive --cql-features normal \
---max-mutation-retries 5 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms \
---replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}\" "
+user_prefix: 'gemini-cdc-preimage-write'
 
+gemini_cmd: |
+  --duration 3h
+  --concurrency 30
+  --mode write
 
 gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
 gemini_table_options:
   - "cdc={'enabled': true, 'preimage': true}"
+gemini_log_cql_statements: false
 
 db_type: scylla
diff --git a/test-cases/gemini/gemini-3h-cdc-write.yaml b/test-cases/gemini/gemini-3h-cdc-write.yaml
@@ -4,18 +4,16 @@ n_loaders: 1
 n_monitor_nodes: 1
 instance_type_db: 'i4i.4xlarge'
 
-user_prefix: "gemini-cdc-write"
+user_prefix: 'gemini-cdc-write'
 
-gemini_cmd: "gemini -d --duration 3h \
--c 30 -m write -f --non-interactive --cql-features normal \
---max-mutation-retries 5 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms \
---replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}\" "
+gemini_cmd: |
+  --duration 3h
+  --concurrency 30
+  --mode write
 
+gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
 gemini_table_options:
   - "cdc={'enabled': true}"
-
-
-gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
+gemini_log_cql_statements: false
 
 db_type: scylla
diff --git a/test-cases/gemini/gemini-3h-ics-cdc-with-nemesis.yaml b/test-cases/gemini/gemini-3h-ics-cdc-with-nemesis.yaml
@@ -11,21 +11,17 @@ ami_db_scylla_user: 'centos'
 nemesis_class_name: 'GeminiChaosMonkey'
 nemesis_interval: 5
 
-# gemini
-# cmd: gemini -d -n [NUM_OF_TEST_ITERATIONS] -c [NUM_OF_THREADS] -m mixed -f
-# the below cmd runs about 3 hours
-gemini_cmd: "gemini -d --duration 3h --warmup 30m -c 50 -m mixed -f --non-interactive \
---cql-features normal --table-options \"compaction={'class': 'IncrementalCompactionStrategy'}\" \
---max-mutation-retries 5 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms \
---replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '3'}\" \
---oracle-replication-strategy \"{'class': 'NetworkTopologyStrategy', 'replication_factor': '1'}\" "
-
+gemini_cmd: |
+  --duration 3h
+  --warmup 30m
+  --concurrency 50
+  --mode write
 
 gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
-
 gemini_table_options:
   - "cdc={'enabled': true, 'preimage': true, 'postimage': true}"
+  - "compaction={'class': 'IncrementalCompactionStrategy'}"
+gemini_log_cql_statements: false
 
 stress_cdclog_reader_cmd: "cdc-stressor -duration 215m -stream-query-round-duration 30s"
 

diff --git a/test-cases/gemini/gemini-3h-ics-with-nondisruptive-nemesis.yaml b/test-cases/gemini/gemini-3h-ics-with-nondisruptive-nemesis.yaml
@@ -10,16 +10,17 @@ ami_db_scylla_user: 'centos'
 
 nemesis_class_name: 'GeminiNonDisruptiveChaosMonkey'
 nemesis_interval: 5
-# gemini
-# cmd: gemini -d -n [NUM_OF_TEST_ITERATIONS] -c [NUM_OF_THREADS] -m mixed -f
-# the below cmd runs about 3 hours
-gemini_cmd: "gemini -d --duration 10800s --warmup 1800s -c 50 -m mixed -f --non-interactive \
---cql-features normal --table-options \"compaction={'class': 'IncrementalCompactionStrategy'}\" \
---max-mutation-retries 5 --max-mutation-retries-backoff 500ms \
---async-objects-stabilization-attempts 5 --async-objects-stabilization-backoff 500ms "
 
+gemini_cmd: |
+  --duration 3h
+  --warmup 30m
+  --concurrency 50
+  --mode write
 
 gemini_schema_url: 'https://s3.amazonaws.com/scylla-gemini/Binaries/schema.json' # currently is not used
+gemini_table_options:
+  - "compaction={'class': 'IncrementalCompactionStrategy'}"
+gemini_log_cql_statements: false
 
 db_type: mixed_scylla
 instance_type_db_oracle: 'i4i.8xlarge'