| Re: Suggestions on the use of lockfiles? [message #33919] |
Sun, 09 February 2003 22:16  |
Jonathan Greenberg
Messages: 91
Registered: November 2002
|
Member |
|
|
Robert:
Thanks! That sounds fine, and is almost what I had come up with
(incidentally, I finally found out the term for this is a "race condition",
since I'm not a programmer it took some searching so I could find an
official term for what was happening) -- here's the only possible issue that
I was concerned about:
If two computers openw,/APPEND and printf the same file, let's say their ID
(computer #1 = 100 and #2 = 200), do we only have 2 possibilities in IDL:
Lockfile contains:
100
200
Or
200
100
Or is it possible to have both writing at the same time, giving some
corrupted ID like:
210000 (200 with "100" inside of it)?
If the corrupted id is possible, it makes the situation a bit more
confusing, since the lockfile needs to be deleted before it can move on, an
this brings up some other issues (who gets to delete the file?)
--j
On 2/9/03 12:08 PM, in article Gyy1a.7580$yn1.672904@twister.austin.rr.com,
"Robert Moss" <rmoss4@houston.rr.com> wrote:
> There is probably 100 different ways to approach this, but I prefer to
> start simple. Try this: when the lockfile is created, write some unique
> identifier into the lock file (e.g. the host name of the machine doing
> the locking). Just before you actually start writing the DB, confirm
> that the first line in the lockfile corresponds to the machine that is
> getting ready to alter the DB. If it does not, it can go back to the
> wait loop.
>
> Robert Moss, PhD
> rmoss4@houston.rr.com
>
> Jonathan Greenberg wrote:
>> I am having some troubles using lockfiles to do simple database read/writes
>> without getting a database corruption (I'm trying to hack my way around
>> buying Dataminer):
>>
>> I have two or more computers running this code:
>>
>> dblockname is the name of the lockfile for the database
>>
>> ***
>>
>> while [program is still running] do begin
>> ; Waits for the lockfile to not exist
>> while (file_test(dblockname) eq 1) do begin
>> wait,1
>> endwhile
>> ; Creates a lockfile
>> openw,dblocklun,dblockname,/GET_LUN
>> printf,dblocklun,''
>> free_lun,dblocklun
>>
>> [gets the database, reads and writes to it]
>> ; Deletes the lockfile, and returns to the top of the loop
>> file_delete,dblockname
>>
>> endwhile
>>
>> ***
>>
>> The problem I'm having is that from time to time (depending on the speed of
>> the database I/O), I'm getting errors where, and this is a guess, one
>> computer manages to check for a nonexistant lockfile at the same time as
>> another, and so both computers break out of the loop at the same time and
>> start fooling around with the database at the same time. At the end,
>> whichever computer deletes the lockfile first can reenter the loop, but the
>> 2nd computer gets an error since the file was already deleted (and the
>> database is likely to be corrupted at this point).
>>
>> Ideas? The astronomy lockfile procedures appear to be unix only, and I'm
>> trying to design this for any platform (Windows, Mac, UNIX). Apparently I
>> can openw the same file from two computers without error (I wish that I
>> could override this fact). Help!
>>
>> --j
>>
>
|
|
|
|
| Re: Suggestions on the use of lockfiles? [message #33922 is a reply to message #33919] |
Sun, 09 February 2003 12:08  |
rmoss4
Messages: 21
Registered: October 2002
|
Junior Member |
|
|
There is probably 100 different ways to approach this, but I prefer to
start simple. Try this: when the lockfile is created, write some unique
identifier into the lock file (e.g. the host name of the machine doing
the locking). Just before you actually start writing the DB, confirm
that the first line in the lockfile corresponds to the machine that is
getting ready to alter the DB. If it does not, it can go back to the
wait loop.
Robert Moss, PhD
rmoss4@houston.rr.com
Jonathan Greenberg wrote:
> I am having some troubles using lockfiles to do simple database read/writes
> without getting a database corruption (I'm trying to hack my way around
> buying Dataminer):
>
> I have two or more computers running this code:
>
> dblockname is the name of the lockfile for the database
>
> ***
>
> while [program is still running] do begin
> ; Waits for the lockfile to not exist
> while (file_test(dblockname) eq 1) do begin
> wait,1
> endwhile
> ; Creates a lockfile
> openw,dblocklun,dblockname,/GET_LUN
> printf,dblocklun,''
> free_lun,dblocklun
>
> [gets the database, reads and writes to it]
> ; Deletes the lockfile, and returns to the top of the loop
> file_delete,dblockname
>
> endwhile
>
> ***
>
> The problem I'm having is that from time to time (depending on the speed of
> the database I/O), I'm getting errors where, and this is a guess, one
> computer manages to check for a nonexistant lockfile at the same time as
> another, and so both computers break out of the loop at the same time and
> start fooling around with the database at the same time. At the end,
> whichever computer deletes the lockfile first can reenter the loop, but the
> 2nd computer gets an error since the file was already deleted (and the
> database is likely to be corrupted at this point).
>
> Ideas? The astronomy lockfile procedures appear to be unix only, and I'm
> trying to design this for any platform (Windows, Mac, UNIX). Apparently I
> can openw the same file from two computers without error (I wish that I
> could override this fact). Help!
>
> --j
>
|
|
|
|
| Re: Suggestions on the use of lockfiles? [message #34009 is a reply to message #33919] |
Mon, 10 February 2003 20:01 |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
Jonathan Greenberg <greenberg@ucdavis.edu> writes:
> Robert:
>
> Thanks! That sounds fine, and is almost what I had come up with
> (incidentally, I finally found out the term for this is a "race condition",
> since I'm not a programmer it took some searching so I could find an
> official term for what was happening) -- here's the only possible issue that
> I was concerned about:
>
> If two computers openw,/APPEND and printf the same file, let's say their ID
> (computer #1 = 100 and #2 = 200), do we only have 2 possibilities in IDL:
... examples of two processes overwriting each other...
What you need to do is make an atomic operation. One way to do that
is write the data to a temporary file, then use rename or "mv" to
rename the temp file to the lockfile name. Since "mv" should be
atomic, there will be no race condition, and the file will be moved
complete. Then proceed from there.
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@cow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
