|
Sawmill Discussion Forum
ashp

unregistered user
|
Jul-05-05, 11:50 PM (PDT) |
|
|
"Nan%"
| |
I posted about this before but never got a reply. When trying to do sessions stuff, I get stuck at: 3. Generating report (1) nan% 00:00:00 00:00:45 What is nan%? I don't know what kind of QA goes on, but I seem to encounter a new and irritating bug in each new release, and I'm getting fed up of it. |
|
|
|
Printer-friendly page | Top |
|
|
- RE: Nan%,
ashp
, Jul-06-05, 00:15 AM, (1)
- RE: Nan%,
ferrar
, Jul-07-05, 04:46 PM, (2)
- RE: Nan%,
ferrar
, Jul-07-05, 04:49 PM, (3)
- RE: Nan%,
ashp
, Jul-08-05, 02:06 AM, (4)
- RE: Nan%,
ashp
, Jul-08-05, 03:16 AM, (5)
- RE: Nan%,
ferrar
, Jul-08-05, 04:49 PM, (6)
- RE: Nan%,
ferrar
, Jul-08-05, 04:53 PM, (7)
|
ashp

unregistered user
|
Jul-06-05, 00:15 AM (PDT) |
|
|
1. "RE: Nan%"
In response to message #0
| |
Thinking further to this, do you guys make pre-releases available for testing? We've stripped this customer back to the start of June so we can fit the session stuff in memory, and assuming we get past the nan% issue that'll help. We're looking to move them to a dedicated box, with another 25profile licence, but obviously they are unconvinced this'll work as a result of all the problems we've had, so it would be nice to be able to test with a pre-release to assure them that 7.2 will do the right thing. |
|
|
|
Printer-friendly page | Top |
|
|
 |
ferrar
Member since Sep-5-01
3687 posts |
Jul-07-05, 04:46 PM (PDT) |
 |
2. "RE: Nan%"
In response to message #1
 |
You *can* download a 7.2 pre-release ( http://sawmill.net/prerelease.html ), but that's definitely not the place to look for greater stability! 7.2 is currently in mid-cycle; it's months from release, so we think nothing of yanking out huge chunks of code and replacing them with new, untested stuff, which will supposedly be much better when it's working, but currently doesn't work all that well. It's very much the bleeding edge of development, and cannot be called a "beta" -- it's not even an "alpha". Still, grab it if you'd like to have some fun and don't mind crashes and wrong answers.... But don't bother (yet) if you're using MySQL--it's not working a bit in 7.2 at the moment. Eventually, when 7.2 calms down, we will have a true beta. That won't be as stable as 7.1 is now, but it will be more stable than 7.2 is now. We'll send mail to the newsletter if the beta goes public. You're not alone in your "one bug after another" problems with 7.1.x. We've had a hard time of it with 7.1.x--we're not adding anything new to the branch other than bug fixes, which should make it more and more stable, and that is happening overall, but we've had a number of cases where one bug fix breaks something else, and our QA scripts don't catch it. We do pretty extensive manual QA, but it hasn't been testing these very well either, so we've really beefed up our automated QA suite lately (in the past couple months), bringing it up from less than ten "sanity check" tests (rebuild a database and check the Overview numbers; rebuild/update/expire and check the Overview; export a couple reports as CSV and check the numbers) to more than 130 tests in the current suite, which build databases using both "internal" and "MySQL", using several type of log data in several charsets, and perform and validate CVS and HTML exports of dozens of reports, with simple filters, date filters, and regular expression filters. 7.1.11, which is about to go out the door, passes all these tests. It took us a couple days of debugging just now to fix a few issues the newest tests turned up, so they are definitely helping. 7.1.11 also had a particularly complete manual QA test (humans at web browsers, trying to break it). I think 7.1.10 was pretty good (it was the first one to really benefit from the new huge test suite), and 7.1.11 will be better. So we've done a lot lately to address stability problems with 7.1, and we'll keep working to make it as bug free as possible. |
|
|
|
Printer-friendly page | Top |
|
|
|
 |
ferrar
Member since Sep-5-01
3687 posts |
Jul-07-05, 04:49 PM (PDT) |
 |
3. "RE: Nan%"
In response to message #2
 |
Regarding NaN (which means "not a number"), that's probably a cosmetic thing; there shouldn't be a NaN there, but it's probably not the problem. The problem is probably that the session analysis is still using too much memory. If that's the case, you may be happy to hear that one of the new features in 7.2 is that the session calculation is done using disk files, and uses very little memory, so "not enough memory for session calculation" should no longer be a problem in 7.2. Many other things which are currently held in memory will also be moved to disk in 7.2 (database itemnum tables, xref tables, subitem lists, uniques lists, and intermediate report tables). These are all implemented already, and recently, in the current 7.2, which is why it's so unstable now, but if you want to give it a whirl, you can live on the edge a little and maybe you'll find that it works for you and lets you process much more data in the same RAM. |
|
|
|
Printer-friendly page | Top |
|
|
|
 |
ashp

unregistered user
|
Jul-08-05, 02:06 AM (PDT) |
|
|
4. "RE: Nan%"
In response to message #3
| |
>Regarding NaN (which means "not a number"), that's probably >a cosmetic thing; there shouldn't be a NaN there, but it's >probably not the problem. The problem is probably that the >session analysis is still using too much memory. If that's >the case, you may be happy to hear that one of the new >features in 7.2 is that the session calculation is done >using disk files, and uses very little memory, so "not >enough memory for session calculation" should no longer be a >problem in 7.2. Many other things which are currently held >in memory will also be moved to disk in 7.2 (database >itemnum tables, xref tables, subitem lists, uniques lists, >and intermediate report tables). These are all implemented >already, and recently, in the current 7.2, which is why it's >so unstable now, but if you want to give it a whirl, you can >live on the edge a little and maybe you'll find that it >works for you and lets you process much more data in the >same RAM.Thanks for the update on this, I appreciate it. Is there a rough limit on how big the database can be before you start running out of memory to do sessions. We've reduced the database from 250G to 24G so far, but it still stalls. Are there any database or filters we can set to drop information to help it? Maybe drop individual IPs so it just works on /24's, or something else along those lines. They mostly just use it for reporting to the board, so I think if I can reduce the detail it will still be effective enough.
|
|
|
|
Printer-friendly page | Top |
|
|
|
 |
ashp

unregistered user
|
Jul-08-05, 03:16 AM (PDT) |
|
|
5. "RE: Nan%"
In response to message #4
| |
Further to this, I did some poking around, and I see the referrers field in items/ is at 333M, so I assume this may account for a large part of the database. I'm trying to find out how to drop fields, I can't seem to do it through the interface however. Do I just have to modify the jessops.cfg and attempt to strip out all references to the referrer, or am I missing something easier? |
|
|
|
Printer-friendly page | Top |
|
|
|
 |
|
 |
ferrar
Member since Sep-5-01
3687 posts |
Jul-08-05, 04:53 PM (PDT) |
 |
7. "RE: Nan%"
In response to message #4
 |
24G is still much to large to do a full session analysis, using the "internal" database. I would say that 2G is around where session analyses become unmanageable with the internal database. Much larger analyses are possible with MySQL (there's no particular limit, if you're willing to wait for the analysis). It is in fact possible to do a session analysis of a 24G database, just not the *whole* database. If you filter first, then the session analysis is done only for the filter set. For instance, if you have a 365GB database which covers a year, and you zoom on one day, then you're really only looking at 1GB worth of data. Do a session analysis on that, and it should work. Remove the filter, and you'll need several hundred GB of RAM to do that analysis.... If you do need a full analysis, then what you're discussing is just the way to proceed--you need to prune your data, using log filters. Drop most IPs if you can; delete references to sites or folders or anything else that you don't care about. |
|
|
|
Printer-friendly page | Top |
|
|
|
|
|