-
Type:
Task
-
Resolution: Done
-
Affects Version/s: None
-
Component/s: None
When running the test1 configuration with 8 tables instead of 1, we see frequent dropouts where throughput is 0. Looking at pmp output, we're waiting on the schema lock which is held by one of the checkpoint worker threads:
20 sched_yield,__wt_yield,__clsm_enter,__clsm_insert,populate_thread,start_thread,clone
13 sched_yield,__wt_yield,__wt_session_get_btree,__wt_session_get_btree_ckpt,__wt_curfile_open,__wt_open_cursor,__bloom_open_cursor,__wt_bloom_hash_get,__wt_bloom_get,__lsm_bloom_create,__lsm_bloom_work,__wt_lsm_merge_worker,start_thread,clone
6 sched_yield,__wt_yield,__wt_clsm_open,__wt_open_cursor,__lsm_bloom_create,__lsm_bloom_work,__wt_lsm_merge_worker,start_thread,clone
5 sched_yield,__wt_yield,__wt_lsm_checkpoint_worker,start_thread,clone
2 sched_yield,__wt_yield,__wt_clsm_open,__wt_open_cursor,__wt_lsm_merge,__wt_lsm_merge_worker,start_thread,clone
1 sched_yield,__wt_yield,__wt_session_get_btree,__wt_session_get_btree_ckpt,__wt_curfile_open,__wt_open_cursor,__wt_lsm_merge,__wt_lsm_merge_worker,start_thread,clone
1 sched_yield,__wt_yield,__wt_session_get_btree,__lsm_discard_handle,__wt_lsm_checkpoint_worker,start_thread,clone
1 sched_yield,__wt_yield,__wt_session_get_btree,__lsm_discard_handle,__lsm_drop_file,__lsm_free_chunks,__wt_lsm_merge_worker,start_thread,clone
1 sched_yield,__wt_yield,__wt_lsm_merge_worker,start_thread,clone
1 pthread_rwlock_rdlock,__wt_readlock,__wt_session_lock_btree,__wt_session_get_btree,__wt_conn_btree_apply,__curstat_file_init,__wt_curstat_init,__wt_curstat_open,__lsm_stat_init,__wt_curstat_lsm_init,__wt_curstat_init,__wt_curstat_open,__statlog_dump,__statlog_lsm_apply,__statlog_server,start_thread,clone
1 pthread_cond_timedwait@@GLIBC_2.3.2,__wt_cond_wait,__wt_lsm_checkpoint_worker,start_thread,clone
1 pthread_cond_timedwait@@GLIBC_2.3.2,__wt_cond_wait,__wt_cache_evict_server,start_thread,clone
1 nanosleep,usleep,execute_populate,start_run,start_all_runs,main
1 nanosleep,sleep,monitor,start_thread,clone
1 __free_skip_list,__free_skip_array,__free_page_row_leaf,__wt_page_out,__wt_evict_file,__wt_bt_cache_op,__checkpoint_worker,__wt_checkpoint_close,__wt_conn_btree_sync_and_close,__wt_session_release_btree,__lsm_discard_handle,__wt_lsm_checkpoint_worker,start_thread,clone
1
Similar stacks exist during all the times we have dropouts.