8.4 KiB
8.4 KiB
# How to execute scenarios
## Common options for gRPC, HTTP, S3 scenarios:
Scenarios
grpc.js, http.js and s3.js support the following options:
* DURATION - duration of scenario in seconds.
* READERS - number of VUs performing read operations.
* WRITERS - number of VUs performing write operations.
* REGISTRY_FILE - if set, all produced objects will be stored in database for subsequent verification. Database file name will be set to the value of REGISTRY_FILE.
* WRITE_OBJ_SIZE - object size in kb for write(PUT) operations.
* PREGEN_JSON - path to json file with pre-generated containers and objects (in case of http scenario we use json pre-generated for grpc scenario).
* SLEEP_WRITE - time interval (in seconds) between writing VU iterations.
* SLEEP_READ - time interval (in seconds) between reading VU iterations.
* SELECTION_SIZE - size of batch to select for deletion (default: 1000).
Examples of how to use these options are provided below for each scenario.
## gRPC
1. Create pre-generated containers or objects:
The tests will use all pre-created containers for PUT operations and all pre-created objects for READ operations.
shell $ ./scenarios/preset/preset_grpc.py --size 1024 --containers 1 --out grpc.json --endpoint host1:8080 --preload_obj 500 --policy "REP 2 IN X CBF 1 SELECT 2 FROM * AS X"
* --policy - container policy. If parameter is omitted, the default value is "REP 1 IN X CBF 1 SELECT 1 FROM * AS X".
* --update - container id. Specify the existing container id, if parameter is omitted the new container will be created.
2. Execute scenario with options:
shell $ ./k6 run -e DURATION=60 -e WRITE_OBJ_SIZE=8192 -e READERS=20 -e WRITERS=20 -e DELETERS=30 -e DELETE_AGE=10 -e REGISTRY_FILE=registry.bolt -e GRPC_ENDPOINTS=host1:8080,host2:8080 -e PREGEN_JSON=./grpc.json scenarios/grpc.js
Options (in addition to the common options):
* GRPC_ENDPOINTS - GRPC endpoints of FrostFS storage in format host:port. To specify multiple endpoints separate them by comma.
* DELETERS - number of VUs performing delete operations (using deleters requires that options DELETE_AGE and REGISTRY_FILE are specified as well).
* DELETE_AGE - age of object in seconds before which it can not be deleted. This parameter can be used to control how many objects we have in the system under load.
* SLEEP_DELETE - time interval (in seconds) between deleting VU iterations.
* DIAL_TIMEOUT - timeout to connect to a node (in seconds).
* STREAM_TIMEOUT - timeout for a single stream message for PUT/GET operations (in seconds).
## HTTP
1. Create pre-generated containers or objects:
There is no dedicated script to preset HTTP scenario, so we use the same script as for gRPC:
shell $ ./scenarios/preset/preset_grpc.py --size 1024 --containers 1 --out grpc.json --endpoint host1:8080 --preload_obj 500
2. Execute scenario with options:
shell $ ./k6 run -e DURATION=60 -e WRITE_OBJ_SIZE=8192 -e READERS=10 -e WRITERS=20 -e REGISTRY_FILE=registry.bolt -e HTTP_ENDPOINTS=host1:8888,host2:8888 -e PREGEN_JSON=./grpc.json scenarios/http.js
Options (in addition to the common options):
* HTTP_ENDPOINTS - endpoints of HTTP gateways in format host:port. To specify multiple endpoints separate them by comma.
## S3
1. Create s3 credentials:
shell $ frostfs-s3-authmate issue-secret --wallet wallet.json --peer host1:8080 --gate-public-key 03d33a2cc7b8daaa5a3df3fccf065f7cf1fc6a3279efc161fcec512dcc0c1b2277 --gate-public-key 03ff0ad212e10683234442530bfd71d0bb18c3fbd6459aba768eacf158b0c359a2 --gate-public-key 033ae03ff30ed3b6665af69955562cfc0eae18d50e798ab31f054ee22e32fee993 --gate-public-key 02127c7498de0765d2461577c9d4f13f916eefd1884896183e6de0d9a85d17f2fb --bearer-rules rules.json --container-placement-policy "REP 1 IN X CBF 1 SELECT 1 FROM * AS X" --container-policy ./scenarios/files/policy.json Enter password for wallet.json > { "access_key_id": "38xRsCTb2LTeCWNK1x5dPYeWC1X22Lq4ahKkj1NV6tPk0Dack8FteJHQaW4jkGWoQBGQ8R8UW6CdoAr7oiwS7fFQb", "secret_access_key": "e671e353375030da3fbf521028cb43810280b814f97c35672484e303037ea1ab", "owner_private_key": "48e83ab313ca45fe73c7489565d55652a822ef659c75eaba2d912449713f8e58", "container_id": "38xRsCTb2LTeCWNK1x5dPYeWC1X22Lq4ahKkj1NV6tPk" }
Run aws configure.
2. Create pre-generated buckets or objects:
The tests will use all pre-created buckets for PUT operations and all pre-created objects for READ operations.
shell $ ./scenarios/preset/preset_s3.py --size 1024 --buckets 1 --out s3_1024kb.json --endpoint host1:8084 --preload_obj 500 --location load-1-4
* '--location' - specify the name of container policy (from policy.json file). It's important to run 'aws configure' each time when the policy file has been changed to pick up the latest policies.
3. Execute scenario with options:
shell $ ./k6 run -e DURATION=60 -e WRITE_OBJ_SIZE=8192 -e READERS=20 -e WRITERS=20 -e DELETERS=30 -e DELETE_AGE=10 -e S3_ENDPOINTS=host1:8084,host2:8084 -e PREGEN_JSON=s3.json scenarios/s3.js
Options (in addition to the common options):
* S3_ENDPOINTS - endpoints of S3 gateways in format host:port. To specify multiple endpoints separate them by comma.
* DELETERS - number of VUs performing delete operations (using deleters requires that options DELETE_AGE and REGISTRY_FILE are specified as well).
* DELETE_AGE - age of object in seconds before which it can not be deleted. This parameter can be used to control how many objects we have in the system under load.
* SLEEP_DELETE - time interval (in seconds) between deleting VU iterations.
* OBJ_NAME - if specified, this name will be used for all write operations instead of random generation.
## Verify
This scenario allows to verify that objects created by a previous run are really stored in the system and their data is not corrupted. Running this scenario assumes that you've already run gRPC or HTTP or S3 scenario with option REGISTRY_FILE.
To verify stored objects execute scenario with options:
./k6 run -e CLIENTS=200 -e TIME_LIMIT=120 -e GRPC_ENDPOINTS=host1:8080,host2:8080 -e S3_ENDPOINTS=host1:8084,host2:8084 -e REGISTRY_FILE=registry.bolt scenarios/verify.js
Scenario picks up all objects in created status. If object is stored correctly, its' status will be changed into verified. If object does not exist or its' data is corrupted, then the status will be changed into invalid.
Scenario ends as soon as all objects are checked or time limit is exceeded.
Running VERIFY scenario modifies status of objects in REGISTRY_FILE. Objects that have been verified once won't be verified again. If you would like to verify the same set of objects multiple times, you can create a copy of REGISTRY_FILE produced by the LOAD scenario and run VERIFY against the copy of the file.
Objects produced by HTTP scenario will be verified via gRPC endpoints.
Options:
* CLIENTS - number of VUs for verifying objects (VU can handle both GRPC and S3 objects)
* TIME_LIMIT - amount of time in seconds that is sufficient to verify all objects. If this time interval ends, then verification process will be interrupted and objects that have not been checked will stay in the created state.
* REGISTRY_FILE - database file from which objects for verification should be read.
* SLEEP - time interval (in seconds) between VU iterations.
* SELECTION_SIZE - size of batch to select for deletion (default: 1000).
* DIAL_TIMEOUT - timeout to connect to a node (in seconds).
* STREAM_TIMEOUT - timeout for a single stream message for PUT/GET operations (in seconds).
## Verify preset
Check what all preset objects is in cluster and can be got:
./scenarios/preset/check_objects_in_preset.py --endpoint az:8080 --preset_file ./scenarios/presets/grpc_1Mb_c1_o100.json
Options:
* --endpoint - endpoint to get objects
* --preset_file - path to preset file
Check what all objects in preset is compliance with container policy and get distribution of keys:
./scenarios/preset/check_policy_compliance.py --endpoints "az:8080,buky:8080,vedi:8080,glagoli:8080" --expected_copies 2 --preset_file "./scenarios/presets/grpc_10Mb_c100_o400.json"
Options:
* --endpoints - list of all live endpoints in cluster (comma separated)
* --preset_file - path to preset file
* --expected_copies - amount of expected copies for each object
* --max_workers - amount of workers for check in parallel
* --print_failed - print failed objects to console