Project

General

Profile

ADMINISTRATION » History » Version 1

Arthur Kreymer, 11/06/2012 12:06 PM

1 1 Arthur Kreymer
h1. ADMINISTRATION
2 1 Arthur Kreymer
3 1 Arthur Kreymer
Working directories and files are under /grid/data/${GROUP}/LOCK
4 1 Arthur Kreymer
5 1 Arthur Kreymer
On the grid, the username does not reflect the identity of the
6 1 Arthur Kreymer
person who submitted the job.
7 1 Arthur Kreymer
So the lock script gets the identity from the grid proxy.
8 1 Arthur Kreymer
9 1 Arthur Kreymer
/LOCKS - active lock files
10 1 Arthur Kreymer
  The lock files are empty, with names contining
11 1 Arthur Kreymer
  date, time queued,               host, pid, user, identity
12 1 Arthur Kreymer
13 1 Arthur Kreymer
/QUEUE - locks pending, empty files containing
14 1 Arthur Kreymer
  date,                            host, pid, user, identity
15 1 Arthur Kreymer
16 1 Arthur Kreymer
/LOG   - empty files with names reflecting completed locks
17 1 Arthur Kreymer
   date, time queued, time locked, host, pid,  user, identity
18 1 Arthur Kreymer
19 1 Arthur Kreymer
/LOGS   - monthly text summaries built from LOG file names.
20 1 Arthur Kreymer
21 1 Arthur Kreymer
/STALE - record of locks that have timed out
22 1 Arthur Kreymer
23 1 Arthur Kreymer
glimit - global activity limit, including all user groups
24 1 Arthur Kreymer
         set this near the actual Bluearc capacity
25 1 Arthur Kreymer
         this is not implemented as of 2012-11-06
26 1 Arthur Kreymer
27 1 Arthur Kreymer
limit  - local activity limit, for the users' own group
28 1 Arthur Kreymer
             set this well under Bluearc capacity
29 1 Arthur Kreymer
30 1 Arthur Kreymer
perf   - performance MB/sec required in PERF before locking
31 1 Arthur Kreymer
32 1 Arthur Kreymer
PERF   - actual MB/sec performance, measured by external agent
33 1 Arthur Kreymer
            ( No agents implemented as of 2010-08-02 )
34 1 Arthur Kreymer
35 1 Arthur Kreymer
rate   - net retry rate target, in retries per second
36 1 Arthur Kreymer
37 1 Arthur Kreymer
small  - MBytes: files smaller than this are not locked by cpn.
38 1 Arthur Kreymer
39 1 Arthur Kreymer
wait   - mininum time to wait before retrying, regardless of the load.
40 1 Arthur Kreymer
         the time delay before retrying a lock is the minimum of
41 1 Arthur Kreymer
* wait
42 1 Arthur Kreymer
* (number of queued locks)/rate
43 1 Arthur Kreymer
44 1 Arthur Kreymer
h1. MAINTENANCE
45 1 Arthur Kreymer
46 1 Arthur Kreymer
lock files should be owned by some appropriate group account, like mindata.
47 1 Arthur Kreymer
48 1 Arthur Kreymer
That account should occasionally remove expired locks and queue entries,
49 1 Arthur Kreymer
and concatenate  LOG entries into monthly summary files.
50 1 Arthur Kreymer
51 1 Arthur Kreymer
You can run the lockclean script, which will do this hourly :
52 1 Arthur Kreymer
53 1 Arthur Kreymer
    set nohup ; /grid/fermiapp/common/tools/lockclean &
54 1 Arthur Kreymer
55 1 Arthur Kreymer
Get an idea of activity by counting lines in log files.
56 1 Arthur Kreymer
57 1 Arthur Kreymer
For example, for Minos, 
58 1 Arthur Kreymer
59 1 Arthur Kreymer
  $ wc -l /grid/data/e875/LOCK/LOGS/*.log
60 1 Arthur Kreymer
    9124 /grid/data/e875/LOCK/LOGS/200908.log
61 1 Arthur Kreymer
  140794 /grid/data/e875/LOCK/LOGS/200909.log
62 1 Arthur Kreymer
  181895 /grid/data/e875/LOCK/LOGS/200910.log
63 1 Arthur Kreymer
  196327 /grid/data/e875/LOCK/LOGS/200911.log
64 1 Arthur Kreymer
  125084 /grid/data/e875/LOCK/LOGS/200912.log
65 1 Arthur Kreymer
  272598 /grid/data/e875/LOCK/LOGS/201001.log
66 1 Arthur Kreymer
  284000 /grid/data/e875/LOCK/LOGS/201002.log
67 1 Arthur Kreymer
  275479 /grid/data/e875/LOCK/LOGS/201003.log
68 1 Arthur Kreymer
  354725 /grid/data/e875/LOCK/LOGS/201004.log
69 1 Arthur Kreymer
 1840026 total
70 1 Arthur Kreymer
71 1 Arthur Kreymer
  
72 1 Arthur Kreymer
  $ wc -l /grid/data/e875/LOCK/STALE/LOCKS/*.log
73 1 Arthur Kreymer
$ wc -l /grid/data/e875/LOCK/STALE/QUEUE/*.log