projects / ceph.git / blame
| Commit | Line | Data |
|---|---|---|
| 7c673cae FG |
1 | """ |
| 2 | Resolve stuck peering | |
| 3 | """ | |
| 4 | import logging | |
| 5 | import time | |
| 6 | ||
| 7 | from teuthology import misc as teuthology | |
| e306af50 | 8 | from tasks.util.rados import rados |
| 7c673cae FG |
9 | |
| 10 | log = logging.getLogger(__name__) | |
| 11 | ||
| 12 | def task(ctx, config): | |
| 13 | """ | |
| 14 | Test handling resolve stuck peering | |
| 15 | ||
| 16 | requires 3 osds on a single test node | |
| 17 | """ | |
| 18 | if config is None: | |
| 19 | config = {} | |
| 20 | assert isinstance(config, dict), \ | |
| 21 | 'Resolve stuck peering only accepts a dict for config' | |
| 22 | ||
| 23 | manager = ctx.managers['ceph'] | |
| 24 | ||
| 25 | while len(manager.get_osd_status()['up']) < 3: | |
| 26 | time.sleep(10) | |
| 27 | ||
| 28 | ||
| 29 | manager.wait_for_clean() | |
| 30 | ||
| 31 | dummyfile = '/etc/fstab' | |
| 32 | dummyfile1 = '/etc/resolv.conf' | |
| 33 | ||
| 34 | #create 1 PG pool | |
| 35 | pool='foo' | |
| 36 | log.info('creating pool foo') | |
| 37 | manager.raw_cluster_cmd('osd', 'pool', 'create', '%s' % pool, '1') | |
| 38 | ||
| 39 | #set min_size of the pool to 1 | |
| 40 | #so that we can continue with I/O | |
| 41 | #when 2 osds are down | |
| 42 | manager.set_pool_property(pool, "min_size", 1) | |
| 43 | ||
| 44 | osds = [0, 1, 2] | |
| 45 | ||
| 46 | primary = manager.get_pg_primary('foo', 0) | |
| 47 | log.info("primary osd is %d", primary) | |
| 48 | ||
| 49 | others = list(osds) | |
| 50 | others.remove(primary) | |
| 51 | ||
| 52 | log.info('writing initial objects') | |
| 53 | first_mon = teuthology.get_first_mon(ctx, config) | |
| 9f95a23c | 54 | (mon,) = ctx.cluster.only(first_mon).remotes.keys() |
| 7c673cae FG |
55 | #create few objects |
| 56 | for i in range(100): | |
| 57 | rados(ctx, mon, ['-p', 'foo', 'put', 'existing_%d' % i, dummyfile]) | |
| 58 | ||
| 59 | manager.wait_for_clean() | |
| 60 | ||
| 61 | #kill other osds except primary | |
| 62 | log.info('killing other osds except primary') | |
| 63 | for i in others: | |
| 64 | manager.kill_osd(i) | |
| 65 | for i in others: | |
| 66 | manager.mark_down_osd(i) | |
| 67 | ||
| 68 | ||
| 69 | for i in range(100): | |
| 70 | rados(ctx, mon, ['-p', 'foo', 'put', 'new_%d' % i, dummyfile1]) | |
| 71 | ||
| 72 | #kill primary osd | |
| 73 | manager.kill_osd(primary) | |
| 74 | manager.mark_down_osd(primary) | |
| 75 | ||
| 76 | #revive other 2 osds | |
| 77 | for i in others: | |
| 78 | manager.revive_osd(i) | |
| 79 | ||
| 80 | #make sure that pg is down | |
| 81 | #Assuming pg number for single pg pool will start from 0 | |
| 82 | pgnum=0 | |
| 83 | pgstr = manager.get_pgid(pool, pgnum) | |
| 84 | stats = manager.get_single_pg_stats(pgstr) | |
| 9f95a23c | 85 | print(stats['state']) |
| 7c673cae FG |
86 | |
| 87 | timeout=60 | |
| 88 | start=time.time() | |
| 89 | ||
| 90 | while 'down' not in stats['state']: | |
| 91 | assert time.time() - start < timeout, \ | |
| 92 | 'failed to reach down state before timeout expired' | |
| 93 | stats = manager.get_single_pg_stats(pgstr) | |
| 94 | ||
| 95 | #mark primary as lost | |
| 96 | manager.raw_cluster_cmd('osd', 'lost', '%d' % primary,\ | |
| 97 | '--yes-i-really-mean-it') | |
| 98 | ||
| 99 | ||
| 100 | #expect the pg status to be active+undersized+degraded | |
| 101 | #pg should recover and become active+clean within timeout | |
| 102 | stats = manager.get_single_pg_stats(pgstr) | |
| 9f95a23c | 103 | print(stats['state']) |
| 7c673cae FG |
104 | |
| 105 | timeout=10 | |
| 106 | start=time.time() | |
| 107 | ||
| 108 | while manager.get_num_down(): | |
| 109 | assert time.time() - start < timeout, \ | |
| 110 | 'failed to recover before timeout expired' | |
| 31f18b77 FG |
111 | |
| 112 | manager.revive_osd(primary) |