Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
1.4.0
-
None
-
None
-
None
-
1.6.0-RC2, Hadoop 2.2.0, AWS 20 node cluster
Description
While running the verification map reduce job as part of the continuous ingest test, I noticed the map phase was taking longer than expected. I had run 24 hours of ingest and then verification. There were 2048 tablets and ~32B entries. List scans showed that a lot of mappers were reading from one node. That single tserver was thrashing and had a much lower aggregate read rate than tservers that only had a few mappers reading (like ~35KV/s vs 150KV/s).
Below is the output of listscans
root@test160> listscans
TABLET SERVER | CLIENT | AGE | LAST | STATE | TYPE | USER | TABLE | COLUMNS | AUTHORIZATIONS | TABLET | ITERATORS | ITERATOR OPTIONS
ip-10-1-2-15:9997 | 10.1.2.14:35838 | 2m47s | 5ms |RUNNING |SINGLE | root | ci | [] | |3;121e33;120f25 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42586 | 2h33m | 248ms |RUNNING |SINGLE | root | ci | [] | |3;422d5;421e3d | [] | {}
ip-10-1-2-23:9997 | 10.1.2.18:60511 | 2h53m | 193ms | QUEUED |SINGLE | root | ci | [] | |3;554b7;553c5e | [] | {}
ip-10-1-2-23:9997 | 10.1.2.13:40589 | 2h19m | 246ms | QUEUED |SINGLE | root | ci | [] | |3;7f1e43e;7f0f3 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.25:55164 | 56m18s | 73ms | QUEUED |SINGLE | root | ci | [] | |3;1bf149b;1be238 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42618 | 2h26m | 263ms |RUNNING |SINGLE | root | ci | [] | |3;555a83;554b7 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:60869 | 1h47m | 131ms |RUNNING |SINGLE | root | ci | [] | |3;42e206d;42d2f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.14:59576 | 4m31s | 71ms |RUNNING |SINGLE | root | ci | [] | |3;225a77;224b6be | [] | {}
ip-10-1-2-23:9997 | 10.1.2.27:35342 | 3h1m | 252ms |RUNNING |SINGLE | root | ci | [] | |3;6587a;65789e | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36073 | 2h31m | 131ms | QUEUED |SINGLE | root | ci | [] | |3;41f107;41e1f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.13:40526 | 2h29m | 350ms |RUNNING |SINGLE | root | ci | [] | |3;423c6;422d5 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.18:60560 | 2h37m | 344ms |RUNNING |SINGLE | root | ci | [] | |3;424b6f;423c6 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.29:45044 | 1h17m | 253ms | QUEUED |SINGLE | root | ci | [] | |3;1d0048;1cf14 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.20:48103 | 3h12m | 277ms |RUNNING |SINGLE | root | ci | [] | |3;400f13;4000000000000004 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36053 | 2h33m | 230ms |RUNNING |SINGLE | root | ci | [] | |3;28b4f;28a5e2 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.15:39470 | 2h57m | 269ms |RUNNING |SINGLE | root | ci | [] | |3;2787b;2778a8 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.26:53819 | 3h27m | 449ms |RUNNING |SINGLE | root | ci | [] | |3;430f37;43002ac | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:32894 | 1h9m | 31ms |RUNNING |SINGLE | root | ci | [] | |3;224b6be;223c5c | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36351 | 49m54s | 263ms | QUEUED |SINGLE | root | ci | [] | |3;5eb4fc;5ea5e7a | [] | {}
ip-10-1-2-23:9997 | 10.1.2.20:48227 | 2h46m | 116ms |RUNNING |SINGLE | root | ci | [] | |3;5b4b7;5b3c68e | [] | {}
ip-10-1-2-23:9997 | 10.1.2.15:39676 | 1h57m | 262ms |RUNNING |SINGLE | root | ci | [] | |3;5e96d;5e87bd | [] | {}
ip-10-1-2-23:9997 | 10.1.2.14:58104 | 2h15m | 245ms |RUNNING |SINGLE | root | ci | [] | |3;545a7f;544b7 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.27:35745 | 1h52m | 231ms | QUEUED |SINGLE | root | ci | [] | |3;417895c;41698 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.22:40331 | 2h30m | 192ms |RUNNING |SINGLE | root | ci | [] | |3;54c3d4f;54b4c | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:60923 | 1h32m | 261ms | QUEUED |SINGLE | root | ci | [] | |3;5f004fc;5ef13f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42506 | 2h54m | 117ms | QUEUED |SINGLE | root | ci | [] | |3;404b67;403c5 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.22:40342 | 2h29m | 34ms | QUEUED |SINGLE | root | ci | [] | |3;6bc3f;6bb4e5d | [] | {}
ip-10-1-2-26:9997 | 10.1.2.16:45905 | 21s291ms | 6s841ms | IDLE |SINGLE | root | ci | [] | |3;4396ab;43879c | [] | {}
ip-10-1-2-18:9997 | 10.1.2.26:48600 | 2m2s | 5ms | IDLE |SINGLE | root | ci | [] | |3;4b1e32;4b0f22 | [] | {}
ip-10-1-2-20:9997 | 10.1.2.21:36546 | 2m18s | 7s920ms | IDLE |SINGLE | root | ci | [] | |3;601e91;600f83 | [] | {}
Below is the output ~20 min later.
root@test160> listscans
TABLET SERVER | CLIENT | AGE | LAST | STATE | TYPE | USER | TABLE | COLUMNS | AUTHORIZATIONS | TABLET | ITERATORS | ITERATOR OPTIONS
ip-10-1-2-15:9997 | 10.1.2.14:36125 | 3m10s | 3ms | IDLE |SINGLE | root | ci | [] | |3;1c4ba5a;1c3c9 | [] | {}
ip-10-1-2-14:9997 | 10.1.2.16:35327 | 5m9s | 1ms |RUNNING |SINGLE | root | ci | [] | |3;1b0f58c;1b004c | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42586 | 2h54m | 509ms |RUNNING |SINGLE | root | ci | [] | |3;422d5;421e3d | [] | {}
ip-10-1-2-23:9997 | 10.1.2.13:40589 | 2h40m | 251ms |RUNNING |SINGLE | root | ci | [] | |3;7f1e43e;7f0f3 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.25:55164 | 1h17m | 26ms |RUNNING |SINGLE | root | ci | [] | |3;1bf149b;1be238 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42618 | 2h47m | 455ms |RUNNING |SINGLE | root | ci | [] | |3;555a83;554b7 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:60869 | 2h8m | 352ms |RUNNING |SINGLE | root | ci | [] | |3;42e206d;42d2f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.14:59576 | 25m43s | 112ms |RUNNING |SINGLE | root | ci | [] | |3;225a77;224b6be | [] | {}
ip-10-1-2-23:9997 | 10.1.2.27:35342 | 3h22m | 113ms | QUEUED |SINGLE | root | ci | [] | |3;6587a;65789e | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36073 | 2h52m | 299ms |RUNNING |SINGLE | root | ci | [] | |3;41f107;41e1f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.13:40526 | 2h50m | 71ms |RUNNING |SINGLE | root | ci | [] | |3;423c6;422d5 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.18:60560 | 2h58m | 160ms | QUEUED |SINGLE | root | ci | [] | |3;424b6f;423c6 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.29:45044 | 1h39m | 426ms |RUNNING |SINGLE | root | ci | [] | |3;1d0048;1cf14 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36053 | 2h54m | 184ms | QUEUED |SINGLE | root | ci | [] | |3;28b4f;28a5e2 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.26:53819 | 3h48m | 263ms |RUNNING |SINGLE | root | ci | [] | |3;430f37;43002ac | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:32894 | 1h30m | 163ms | QUEUED |SINGLE | root | ci | [] | |3;224b6be;223c5c | [] | {}
ip-10-1-2-23:9997 | 10.1.2.16:36351 | 1h11m | 180ms |RUNNING |SINGLE | root | ci | [] | |3;5eb4fc;5ea5e7a | [] | {}
ip-10-1-2-23:9997 | 10.1.2.20:48227 | 3h7m | 317ms |RUNNING |SINGLE | root | ci | [] | |3;5b4b7;5b3c68e | [] | {}
ip-10-1-2-23:9997 | 10.1.2.15:39676 | 2h19m | 160ms |RUNNING |SINGLE | root | ci | [] | |3;5e96d;5e87bd | [] | {}
ip-10-1-2-23:9997 | 10.1.2.14:58104 | 2h36m | 238ms |RUNNING |SINGLE | root | ci | [] | |3;545a7f;544b7 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.27:35745 | 2h13m | 162ms |RUNNING |SINGLE | root | ci | [] | |3;417895c;41698 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.22:40331 | 2h51m | 72ms | QUEUED |SINGLE | root | ci | [] | |3;54c3d4f;54b4c | [] | {}
ip-10-1-2-23:9997 | 10.1.2.17:60923 | 1h53m | 27ms | QUEUED |SINGLE | root | ci | [] | |3;5f004fc;5ef13f | [] | {}
ip-10-1-2-23:9997 | 10.1.2.28:42506 | 3h16m | 268ms |RUNNING |SINGLE | root | ci | [] | |3;404b67;403c5 | [] | {}
ip-10-1-2-23:9997 | 10.1.2.22:40342 | 2h51m | 239ms | QUEUED |SINGLE | root | ci | [] | |3;6bc3f;6bb4e5d | [] | {}
ip-10-1-2-29:9997 | 10.1.2.15:56044 | 4s505ms | 3ms | IDLE |SINGLE | root | ci | [] | |3;602da;601e91 | [] | {}
ip-10-1-2-16:9997 | 10.1.2.21:50234 | 51s534ms | 9ms |RUNNING |SINGLE | root | ci | [] | |3;20f12;20e218f | [] | {}
ip-10-1-2-16:9997 | 10.1.2.26:50232 | 3m10s | 5ms |RUNNING |SINGLE | root | ci | [] | |3;5e4b66;5e3c5 | [] | {}
ip-10-1-2-16:9997 | 10.1.2.21:50206 | 3m47s | 285ms | IDLE |SINGLE | root | ci | [] | |3;1ad31;1ac400b | [] | {}
ip-10-1-2-28:9997 | 10.1.2.18:38857 | 42s643ms | 6ms | IDLE |SINGLE | root | ci | [] | |3;2ef13;2ee229 | [] | {}
ip-10-1-2-20:9997 | 10.1.2.20:44062 | 1m23s | 4s928ms | IDLE |SINGLE | root | ci | [] | |3;6b4b7;6b3c669 | [] | {}
I am not sure what caused things to get in this situation, but I have a theory. While the mappers were running a single AWS node was rebooted for some reason. This would have caused tablets to migrate. AccumuloInputFormat calculates its locality information up front, if tablets move mappers will run where the tablets used to be. So maybe a slighty higher than avg number of tablets started reading from ip-10-1-2-23 as a result of the migration. This caused those mapper to run slower and over time more mappers read from ip-10-1-2-23 and things just snowballed.
Regardless of how this situation occurred, Accumulo should handle it better when it does occur. If a single tablet server has much higher number of clients than avg attempting to read for long periods of time, then something should be done. In this case decisions could not be made off of the read rate, because this tserver had a much lower read rate than other tservers that only had 1 or 2 mappers reading.