X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/85f5272570db52619b99f11c879c475743f36aee..f98e0188777b3e2d229c968824b3e64307dae4e6:/services/nodemanager/tests/integration_test.py diff --git a/services/nodemanager/tests/integration_test.py b/services/nodemanager/tests/integration_test.py index 94ac817728..508e626639 100755 --- a/services/nodemanager/tests/integration_test.py +++ b/services/nodemanager/tests/integration_test.py @@ -1,4 +1,8 @@ #!/usr/bin/env python +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: AGPL-3.0 + """Integration test framework for node manager. Runs full node manager with an API server (needs ARVADOS_API_HOST and @@ -21,9 +25,13 @@ from functools import partial import arvados import StringIO +formatter = logging.Formatter('%(asctime)s %(levelname)s: %(message)s') + +handler = logging.StreamHandler(sys.stderr) +handler.setFormatter(formatter) logger = logging.getLogger("logger") logger.setLevel(logging.INFO) -logger.addHandler(logging.StreamHandler(sys.stderr)) +logger.addHandler(handler) detail = logging.getLogger("detail") detail.setLevel(logging.INFO) @@ -31,11 +39,14 @@ if os.environ.get("ANMTEST_LOGLEVEL"): detail_content = sys.stderr else: detail_content = StringIO.StringIO() -detail.addHandler(logging.StreamHandler(detail_content)) +handler = logging.StreamHandler(detail_content) +handler.setFormatter(formatter) +detail.addHandler(handler) fake_slurm = None compute_nodes = None all_jobs = None +unsatisfiable_job_scancelled = None def update_script(path, val): with open(path+"_", "w") as f: @@ -47,16 +58,43 @@ def update_script(path, val): def set_squeue(g): global all_jobs update_script(os.path.join(fake_slurm, "squeue"), "#!/bin/sh\n" + - "\n".join("echo '1|100|100|%s|%s'" % (v, k) for k,v in all_jobs.items())) + "\n".join("echo '1|100|100|%s|%s|(null)|1234567890'" % (v, k) for k,v in all_jobs.items())) return 0 +def set_queue_unsatisfiable(g): + global all_jobs, unsatisfiable_job_scancelled + # Simulate a job requesting a 99 core node. + update_script(os.path.join(fake_slurm, "squeue"), "#!/bin/sh\n" + + "\n".join("echo '99|100|100|%s|%s|(null)|1234567890'" % (v, k) for k,v in all_jobs.items())) + update_script(os.path.join(fake_slurm, "scancel"), "#!/bin/sh\n" + + "\ntouch %s" % unsatisfiable_job_scancelled) + return 0 + +def job_cancelled(g): + global unsatisfiable_job_scancelled + cancelled_job = g.group(1) + api = arvados.api('v1') + # Check that 'scancel' was called + if not os.path.isfile(unsatisfiable_job_scancelled): + return 1 + # Check for the log entry + log_entry = api.logs().list( + filters=[ + ['object_uuid', '=', cancelled_job], + ['event_type', '=', 'stderr'], + ]).execute()['items'][0] + if not re.match( + r"Constraints cannot be satisfied", + log_entry['properties']['text']): + return 1 + return 0 def node_paired(g): global compute_nodes compute_nodes[g.group(1)] = g.group(3) update_script(os.path.join(fake_slurm, "sinfo"), "#!/bin/sh\n" + - "\n".join("echo '%s alloc'" % (v) for k,v in compute_nodes.items())) + "\n".join("echo '%s|alloc|(null)'" % (v) for k,v in compute_nodes.items())) for k,v in all_jobs.items(): if v == "ReqNodeNotAvail": @@ -69,7 +107,7 @@ def node_paired(g): def remaining_jobs(g): update_script(os.path.join(fake_slurm, "sinfo"), "#!/bin/sh\n" + - "\n".join("echo '%s alloc'" % (v) for k,v in compute_nodes.items())) + "\n".join("echo '%s|alloc|(null)'" % (v) for k,v in compute_nodes.items())) for k,v in all_jobs.items(): all_jobs[k] = "Running" @@ -81,7 +119,7 @@ def remaining_jobs(g): def node_busy(g): update_script(os.path.join(fake_slurm, "sinfo"), "#!/bin/sh\n" + - "\n".join("echo '%s idle'" % (v) for k,v in compute_nodes.items())) + "\n".join("echo '%s|idle|(null)'" % (v) for k,v in compute_nodes.items())) return 0 def node_shutdown(g): @@ -109,8 +147,11 @@ def expect_count(count, checks, pattern, g): checks[pattern] = partial(expect_count, count-1) return 0 -def run_test(name, actions, checks, driver_class, jobs): +def run_test(name, actions, checks, driver_class, jobs, provider): code = 0 + global unsatisfiable_job_scancelled + unsatisfiable_job_scancelled = os.path.join(tempfile.mkdtemp(), + "scancel_called") # Delete any stale node records api = arvados.api('v1') @@ -137,7 +178,7 @@ def run_test(name, actions, checks, driver_class, jobs): update_script(os.path.join(fake_slurm, "sinfo"), "#!/bin/sh\n") # Write configuration file for test - with open("tests/fake.cfg.template") as f: + with open("tests/fake_%s.cfg.template" % provider) as f: open(os.path.join(fake_slurm, "id_rsa.pub"), "w").close() with open(os.path.join(fake_slurm, "fake.cfg"), "w") as cfg: cfg.write(f.read().format(host=os.environ["ARVADOS_API_HOST"], @@ -155,7 +196,7 @@ def run_test(name, actions, checks, driver_class, jobs): # Test main loop: # - Read line - # - Apply negative checks (thinks that are not supposed to happen) + # - Apply negative checks (things that are not supposed to happen) # - Check timeout # - Check if the next action should trigger # - If all actions are exhausted, terminate with test success @@ -209,6 +250,7 @@ def run_test(name, actions, checks, driver_class, jobs): code = 1 shutil.rmtree(fake_slurm) + shutil.rmtree(os.path.dirname(unsatisfiable_job_scancelled)) if code == 0: logger.info("%s passed", name) @@ -224,21 +266,46 @@ def main(): # Test lifecycle. tests = { - "test_single_node": ( + "test_unsatisfiable_jobs" : ( + # Actions (pattern -> action) + [ + (r".*Daemon started", set_queue_unsatisfiable), + (r".*Cancelled unsatisfiable job '(\S+)'", job_cancelled), + ], + # Checks (things that shouldn't happen) + { + r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": fail, + r".*Trying to cancel job '(\S+)'": fail, + }, + # Driver class + "arvnodeman.test.fake_driver.FakeDriver", + # Jobs + {"34t0i-dz642-h42bg3hq4bdfpf9": "ReqNodeNotAvail"}, + # Provider + "azure"), + "test_single_node_azure": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Suggesting shutdown because node state is \('idle', 'open', .*\)", noop), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), - ], { + ], + # Checks (things that shouldn't happen) + { r".*Suggesting shutdown because node state is \('down', .*\)": fail, r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 1), r".*Setting node quota.*": fail, }, + # Driver class "arvnodeman.test.fake_driver.FakeDriver", - {"34t0i-dz642-h42bg3hq4bdfpf9": "ReqNodeNotAvail"}), + # Jobs + {"34t0i-dz642-h42bg3hq4bdfpf9": "ReqNodeNotAvail"}, + # Provider + "azure"), "test_multiple_nodes": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), @@ -251,48 +318,56 @@ def main(): (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), - ], { + ], + # Checks (things that shouldn't happen) + { r".*Suggesting shutdown because node state is \('down', .*\)": fail, r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 4), r".*Setting node quota.*": fail, }, + # Driver class "arvnodeman.test.fake_driver.FakeDriver", + # Jobs {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf2": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf3": "ReqNodeNotAvail", - "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail" - }), + "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail"}, + # Provider + "azure"), "test_hit_quota": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), - (r".*setting node quota to 3", noop), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Suggesting shutdown because node state is \('idle', 'open', .*\)", remaining_jobs), - (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown) - ], { + ], + # Checks (things that shouldn't happen) + { r".*Suggesting shutdown because node state is \('down', .*\)": fail, r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 2), r".*Sending create_node request.*": partial(expect_count, 5) }, + # Driver class "arvnodeman.test.fake_driver.QuotaDriver", + # Jobs {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf2": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf3": "ReqNodeNotAvail", - "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail" - }), + "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail"}, + # Provider + "azure"), "test_probe_quota": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), - (r".*setting node quota to 3", noop), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Suggesting shutdown because node state is \('idle', 'open', .*\)", remaining_jobs), - (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*sending request", jobs_req), @@ -306,18 +381,24 @@ def main(): (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), - ], { + ], + # Checks (things that shouldn't happen) + { r".*Suggesting shutdown because node state is \('down', .*\)": fail, r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 6), r".*Sending create_node request.*": partial(expect_count, 9) }, + # Driver class "arvnodeman.test.fake_driver.QuotaDriver", + # Jobs {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf2": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf3": "ReqNodeNotAvail", - "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail" - }), + "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail"}, + # Provider + "azure"), "test_no_hang_failing_node_create": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), (r".*Client error: nope", noop), @@ -325,23 +406,75 @@ def main(): (r".*Client error: nope", noop), (r".*Client error: nope", noop), ], + # Checks (things that shouldn't happen) {}, + # Driver class "arvnodeman.test.fake_driver.FailingDriver", + # Jobs {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf2": "ReqNodeNotAvail", "34t0i-dz642-h42bg3hq4bdfpf3": "ReqNodeNotAvail", - "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail" - }), + "34t0i-dz642-h42bg3hq4bdfpf4": "ReqNodeNotAvail"}, + # Provider + "azure"), "test_retry_create": ( + # Actions (pattern -> action) [ (r".*Daemon started", set_squeue), (r".*Rate limit exceeded - scheduling retry in 12 seconds", noop), + (r".*Rate limit exceeded - scheduling retry in 2 seconds", noop), (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", noop), ], + # Checks (things that shouldn't happen) {}, + # Driver class "arvnodeman.test.fake_driver.RetryDriver", - {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail" - }) + # Jobs + {"34t0i-dz642-h42bg3hq4bdfpf1": "ReqNodeNotAvail"}, + # Provider + "azure"), + "test_single_node_aws": ( + # Actions (pattern -> action) + [ + (r".*Daemon started", set_squeue), + (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), + (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), + (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Suggesting shutdown because node state is \('idle', 'open', .*\)", noop), + (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), + ], + # Checks (things that shouldn't happen) + { + r".*Suggesting shutdown because node state is \('down', .*\)": fail, + r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 1), + r".*Setting node quota.*": fail, + }, + # Driver class + "arvnodeman.test.fake_driver.FakeAwsDriver", + # Jobs + {"34t0i-dz642-h42bg3hq4bdfpf9": "ReqNodeNotAvail"}, + # Provider + "ec2"), + "test_single_node_gce": ( + # Actions (pattern -> action) + [ + (r".*Daemon started", set_squeue), + (r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)", node_paired), + (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Not eligible for shut down because node state is \('busy', 'open', .*\)", node_busy), + (r".*ComputeNodeMonitorActor\..*\.([^[]*).*Suggesting shutdown because node state is \('idle', 'open', .*\)", noop), + (r".*ComputeNodeShutdownActor\..*\.([^[]*).*Shutdown success", node_shutdown), + ], + # Checks (things that shouldn't happen) + { + r".*Suggesting shutdown because node state is \('down', .*\)": fail, + r".*Cloud node (\S+) is now paired with Arvados node (\S+) with hostname (\S+)": partial(expect_count, 1), + r".*Setting node quota.*": fail, + }, + # Driver class + "arvnodeman.test.fake_driver.FakeGceDriver", + # Jobs + {"34t0i-dz642-h42bg3hq4bdfpf9": "ReqNodeNotAvail"}, + # Provider + "gce") } code = 0