Properly programmed locked clauses are exactly the means by which one
always avoids deadlocks.
I can think of no example where that would not work, including your
earlier example.
As an example, I have added LOCKED clauses to the 2 pgms I listed below
W12 & W21.
If you run the original versions and wait for both to get to the prompt
before hitting<enter> to both, you can't help but always get a deadlock.
If you run the LOCKED-clause-enabled versions, you can never get a deadlock.
(Forget about the GOTOs. I wanted to keep it simple to read. You can
do the same with LOOP-REPEAT.)
>CT CDS.BP W12 W21
W12
0001 CRT TIMEDATE(), 'W12'
0002 OPEN 'CDSTMP' TO F ELSE STOP
0003 RDUS: READU REC1 FROM F, 'W1' LOCKED
0004 SLEEP 1 ; * wait a bit
0005 GO RDUS
0006 END ELSE NULL
0007 INPUT DUM
0008 CRT TIMEDATE():'...':
0009 READU REC2 FROM F, 'W2' LOCKED
0010 RELEASE F, 'W1'
0011 SLEEP 1 ; * wait a bit
0012 GO RDUS
0013 END ELSE NULL
0014 WRITE TIME() TO F, 'W2'
0015 CRT TIMEDATE()
W21
0001 CRT TIMEDATE(), 'W21'
0002 OPEN 'CDSTMP' TO F ELSE STOP
0003 RDUS: READU REC1 FROM F, 'W2' LOCKED
0004 SLEEP 1 ; * wait a bit
0005 GO RDUS
0006 END ELSE NULL
0007 INPUT DUM
0008 CRT TIMEDATE():'...':
0009 READU REC2 FROM F, 'W1' LOCKED
0010 RELEASE F, 'W2'
0011 SLEEP 1 ; * wait a bit
0012 GO RDUS
0013 END ELSE NULL
0014 WRITE TIME() TO F, 'W2'
0015 CRT TIMEDATE()
On 10/25/2011 6:34 PM, Wjhonson wrote:
Thanks for that longer explanation, but you can't avoid deadlocks with LOCKED
clauses.
Maybe what you meant was something like "You can try to think of some strategy to
deal with them when they occur."
-----Original Message-----
From: Charles Stevenson<[email protected]>
To: U2 Users List<[email protected]>
Sent: Tue, Oct 25, 2011 4:15 pm
Subject: Re: [U2] [UV] LIST.READU EVERY's "waiters" when there are writes w/o
explicit readu.
English is my 1st language, but that doesn't mean I'm good at it.
et me try again.
1. If either (or both) of these have WRITEs without explicit readus
head of time:
- the deadlock daemon seems to recognize the deadlock.
- If said WRITE command cannot gain a lock within 20 minutes it will
rror out. (That is the default behavior, but configurable.)
- So after 20 minutes (one of) the program(s) waiting on such a write
ill error out allowing the other to get the lock and continue.
2. If both programs have READUs:
- the deadlock deamon sees the condition and kills one of them.
The solution for completely avoiding a deadlock is - as always - to
roperly deploy LOCKED clauses.
n 10/25/2011 5:29 PM, Wjhonson wrote:
I dont' quite follow that.
Are you stating that if there is no explicit READU that both processes will
omplete without the deadlock deamon?
But if there is explicit READUs in both processes, that one of them will fail
n 20 minutes?
-----Original Message-----
From: Charles Stevenson<[email protected]>
To: U2 Users List<[email protected]>
Sent: Tue, Oct 25, 2011 3:12 pm
Subject: Re: [U2] [UV] LIST.READU EVERY's "waiters" when there are writes w/o
xplicit readu.
On 10/24/2011 4:11 PM, Charles Stevenson wrote:
--- [snip] ---
I have not yet explored what the deadlock daemon does.
Deadlock daemon ignores these WRITEs w/o explicit locking before hand.
ran these 2 pgms simultaneously so that they both tried to lock or
rite the lock that the other held.
an with lines 6/7 variously commented out or left in.
W12 W21
01: CRT TIMEDATE(), 'W12' | CRT TIMEDATE(), 'W21'
2: OPEN 'CDSTMP' TO F ELSE STOP | OPEN 'CDSTMP' TO F ELSE STOP
3: READU REC1 FROM F, 'W1' ELSE NULL | READU REC2 FROM F, 'W2' ELSE NULL
4: INPUT DUM | INPUT DUM
5: CRT TIMEDATE():'...': | CRT TIMEDATE():'...':
6: *** READU REC2 FROM F, W2 ELSE NULL | READU REC1 FROM F, 'W1' ELSE NULL
7: WRITE TIME() TO F, 'W2' | WRITE TIME() TO F, 'W1'
8: CRT TIMEDATE() | CRT TIMEDATE()
hey invariably entered a deadly embrace, both processes waiting on the
ther, but the deadlock daemon ignored it if there is no explicit readu.
Instead one of them aborted with a write failure in 20 minutes (the
efault, but configurable action), then the other then continued happily.
hope no one but I ever cares about this.
Moral of the story:
Explicitly lock before you update.
cds
n 10/24/2011 4:11 PM, Charles Stevenson wrote:
UV 10.2.10 on Windows is behaving differently from what I recall.
Are my expectations out of line?
Suppose Session A holds a readu lock; and Session B attempts a WRITE
to same record withOUT!!! 1st explicitly getting the readu lock.
Session B waits for Session A to release the lock before writing the
record.
While Session B is waiting, does it show up as a "waiter" in
LIST.READU EVERY?
I expected so, but it doesn't.
Session A Session B
_____________________________ ___________________
1A. ED VOC DUMMY
(this sets the readu lock.)
2A. (stay in editor) 2B. run this:
01: OPEN 'VOC' TO F ELSE STOPM
'nope'
02: ***READU REC FROM F, 'DUMMY'
ELSE NULL
03: WRITE '' TO F, 'DUMMY'
3A. Within ED:
XEQ LIST.READU EVERY
If I UN-comment line 2, LIST.READU EVERY shows something like this:
Active Read Waiters: Owner Waiter
Device.... Inode.... Userno Userno
746117947 232860913 6116 3396
But when I comment out line 2, LIST.READU is silent.
I have not yet explored what the deadlock daemon does.
TIA,
cds
P.S. Yes, yes, "Bad Form", "Legacy Software", 20 min wait is
configurable, . . . we can talk later.
_______________________________________________
U2-Users mailing list
[email protected]
http://listserver.u2ug.org/mailman/listinfo/u2-users
