src/ceph/qa/tasks/cephfs/test_mantle.py

   1 from tasks.cephfs.cephfs_test_case import CephFSTestCase
   2 import json
   3 import logging
   4
   5 log = logging.getLogger(__name__)
   6 failure = "using old balancer; mantle failed for balancer="
   7 success = "mantle balancer version changed: "
   8
   9 class TestMantle(CephFSTestCase):
  10     def start_mantle(self):
  11         self.wait_for_health_clear(timeout=30)
  12         self.fs.set_max_mds(2)
  13         self.wait_until_equal(lambda: len(self.fs.get_active_names()), 2, 30,
  14                               reject_fn=lambda v: v > 2 or v < 1)
  15
  16         for m in self.fs.get_active_names():
  17             self.fs.mds_asok(['config', 'set', 'debug_objecter', '20'], mds_id=m)
  18             self.fs.mds_asok(['config', 'set', 'debug_ms', '0'], mds_id=m)
  19             self.fs.mds_asok(['config', 'set', 'debug_mds', '0'], mds_id=m)
  20             self.fs.mds_asok(['config', 'set', 'debug_mds_balancer', '5'], mds_id=m)
  21
  22     def push_balancer(self, obj, lua_code, expect):
  23         self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer', obj)
  24         self.fs.rados(["put", obj, "-"], stdin_data=lua_code)
  25         with self.assert_cluster_log(failure + obj + " " + expect):
  26             log.info("run a " + obj + " balancer that expects=" + expect)
  27
  28     def test_version_empty(self):
  29         self.start_mantle()
  30         expect = " : (2) No such file or directory"
  31
  32         ret = self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer')
  33         assert(ret == 22) # EINVAL
  34
  35         self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer', " ")
  36         with self.assert_cluster_log(failure + " " + expect): pass
  37
  38     def test_version_not_in_rados(self):
  39         self.start_mantle()
  40         expect = failure + "ghost.lua : (2) No such file or directory"
  41         self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer', "ghost.lua")
  42         with self.assert_cluster_log(expect): pass
  43
  44     def test_balancer_invalid(self):
  45         self.start_mantle()
  46         expect = ": (22) Invalid argument"
  47
  48         lua_code = "this is invalid lua code!"
  49         self.push_balancer("invalid.lua", lua_code, expect)
  50
  51         lua_code = "BAL_LOG()"
  52         self.push_balancer("invalid_log.lua", lua_code, expect)
  53
  54         lua_code = "BAL_LOG(0)"
  55         self.push_balancer("invalid_log_again.lua", lua_code, expect)
  56
  57     def test_balancer_valid(self):
  58         self.start_mantle()
  59         lua_code = "BAL_LOG(0, \"test\")\nreturn {3, 4}"
  60         self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer', "valid.lua")
  61         self.fs.rados(["put", "valid.lua", "-"], stdin_data=lua_code)
  62         with self.assert_cluster_log(success + "valid.lua"):
  63             log.info("run a valid.lua balancer")
  64
  65     def test_return_invalid(self):
  66         self.start_mantle()
  67         expect = ": (22) Invalid argument"
  68
  69         lua_code = "return \"hello\""
  70         self.push_balancer("string.lua", lua_code, expect)
  71
  72         lua_code = "return 3"
  73         self.push_balancer("number.lua", lua_code, expect)
  74
  75         lua_code = "return {}"
  76         self.push_balancer("dict_empty.lua", lua_code, expect)
  77
  78         lua_code = "return {\"this\", \"is\", \"a\", \"test\"}"
  79         self.push_balancer("dict_of_strings.lua", lua_code, expect)
  80
  81         lua_code = "return {3, \"test\"}"
  82         self.push_balancer("dict_of_mixed.lua", lua_code, expect)
  83
  84         lua_code = "return {3}"
  85         self.push_balancer("not_enough_numbers.lua", lua_code, expect)
  86
  87         lua_code = "return {3, 4, 5, 6, 7, 8, 9}"
  88         self.push_balancer("too_many_numbers.lua", lua_code, expect)
  89
  90     def test_dead_osd(self):
  91         self.start_mantle()
  92         expect = " : (110) Connection timed out"
  93
  94         # kill the OSDs so that the balancer pull from RADOS times out
  95         osd_map = json.loads(self.fs.mon_manager.raw_cluster_cmd('osd', 'dump', '--format=json-pretty'))
  96         for i in range(0, len(osd_map['osds'])):
  97           self.fs.mon_manager.raw_cluster_cmd_result('osd', 'down', str(i))
  98           self.fs.mon_manager.raw_cluster_cmd_result('osd', 'out', str(i))
  99
 100         # trigger a pull from RADOS
 101         self.fs.mon_manager.raw_cluster_cmd_result('fs', 'set', self.fs.name, 'balancer', "valid.lua")
 102
 103         # make the timeout a little longer since dead OSDs spam ceph -w
 104         with self.assert_cluster_log(failure + "valid.lua" + expect, timeout=30):
 105             log.info("run a balancer that should timeout")
 106
 107         # cleanup
 108         for i in range(0, len(osd_map['osds'])):
 109           self.fs.mon_manager.raw_cluster_cmd_result('osd', 'in', str(i))