[HECnet] Tops-20 SETNOD Failure
Thomas DeBellis
tommytimesharing at gmail.com
Tue May 4 21:38:57 PDT 2021
I finished the modifications to SCLINK to properly return error values
which are negative and JNTMAN to return the error value in AC3 if .NDINT
doesn't succeed inserting all the nodes. Then I modified SETNOD to get
this extended error information and print it. I put the new monitor and
SETNOD up, rebooted *…AND*…
SETNOD>set nod 2.298 name REACH SETNOD>ins SETNOD>
It works perfectly because, of course it does…
So, as usual, Johnny's guess is pretty close to the mark, even if he
isn't a 36 bit'er. "Slightly broken"? Yeah, 'slightly' enough so that
it can't be easily reproduced…
The only thing I can think of is that the system had been up over 15
weeks when I saw this. I had looked at the storage space utilization
with SYSDPY and didn't notice anything maxing out. I restarted the
GETNOD batch job on VENTI2::. Maybe in another 15 weeks, it will break
again.
/Annoyed/…
> ------------------------------------------------------------------------
> On 5/4/21 10:31 PM, Thomas DeBellis wrote:
>
> Personally, I don't see how it could /possibly/ be anything to do with
> the REACH:: node definition, but I have been known to occasionally
> overlook the utterly obvious, particularly when it's near night-night.
> Maybe not this time.
>
> Right now, the way to figure it out is to get the minor error data and
> see where that takes things. So I'm making a change to JNTMAN to have
> .NDINT to return the lower level code on an incomplete insert. SCLINK
> appears to have a problem that it is mangling return values, which I'm
> currently investigating.
>
> You can't just blithely assuming somebody got it wrong and 'fix'
> things; sometimes it's a certain way for a reason.
>
> On 5/4/21 8:46 PM, Johnny Billquist wrote:
>> On 2021-05-05 00:54, Mike Kostersitz wrote:
>>> Ouch that is one of my nodes 😊 @Johnny Billquist
>>> <mailto:bqt at softjar.se> anything you could think of since we just
>>> renamed my old RSX11M node to REACH.
>>
>> Well. It is something slightly broken in Tops-20, so there isn't
>> really anything we can do about it.
>>
>> Except hope that Thomas can figure it out and fix it.
>>
>> Johnny
>>
>>>
>>> Mike
>>>
>>> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for
>>> Windows 10
>>>
>>> *From: *Thomas DeBellis <mailto:tommytimesharing at gmail.com>
>>> *Sent: *Tuesday, May 4, 2021 15:16
>>> *To: *HECnet <mailto:hecnet at update.uu.se>
>>> *Subject: *[HECnet] Re: Tops-20 SETNOD Failure
>>>
>>> I fixed a few things in SETNOD to get some more information about
>>> the error. In particular,
>>>
>>> * Allow listing of AREA 1 (this was specifically disallowed, I don't
>>> know why)
>>> * More consistent error reporting (via ESOUT%)
>>> * List more than one node when doing an area list (it would only list
>>> a single node)
>>> * List nodes with more than three digits in the node number when
>>> doing
>>> columnar output
>>>
>>> So now you get the expected results:
>>>
>>> SETNOD>lis a 1
>>> [Area 1]
>>> A1RTR 1023 ATHENA 620 ATLE 605 AURORA 606
>>> BANAI 770
>>> BANX25 771 BEA 19 BIZET 800 BJARNE 7
>>> BLINKY 266
>>> CATWZL 302 CLYDE 269 COOPER 263 CRISPS 201
>>> CYGNUS 259
>>> DAVROS 254 DBIT 351 DE1RSX 450 DE1RSY 452
>>> DOCTOR 252
>>> ELIN 616 ELMER 617 ERNIE 2 ERSATZ 350
>>> FLETCH 100
>>> FNATTE 3 FREJ 608 GAXP 730 GNAT 16
>>> GNOME 6
>>> GOBLIN 4 GVAX 731 HAGMAN 262 HARPER 261
>>> HORSE 150
>>> HUGIN 602 HYUNA 500 INKY 268 JIMIN 501
>>> JOCKE 21
>>> JOSSE 17 KLIO 451 KRILLE 8 LOKE 607
>>> MACARO 303
>>> MACRA 258 MAGICA 1 MASTER 251 MIM 13
>>> MUNIN 603
>>> NIPPER 202 NOMAD 610 NOXBIT 720 ORACLE 301
>>> PACMAN 265
>>> PAI 541 PALLAS 621 PAMINA 18 PIDP11 560
>>> PINKY 267
>>> PISTON 520 PLINTH 200 PMAVS2 510 PONDUS 15
>>> PONY 12
>>> PUFF 22 QEMUNT 151 REI 540 ROCKY 11
>>> ROJIN 542
>>> RSX124 306 RSX145 304 RSX170 305 RSX184 307
>>> RUTAN 255
>>> SHARPE 260 SIDRAT 253 SIGGE 10 SPEEDY 24
>>> TARDIS 250
>>> TEMPO 9 THOROS 257 TINA 14 TIPSY 604
>>> TONGUE 264
>>> TOPSY 601 VALAR 400 VAROS 256 WXP 20
>>> WXP2 23
>>> YMER 609 ZEKE 5
>>> Total nodes in area 1: 92
>>> SETNOD>exit
>>>
>>> Regarding the error, I have reproduced it with a single entry, viz:
>>>
>>> !setnod
>>> SETNOD>_set nod 2.298 name REACH_
>>> SETNOD>_insert_
>>> ?SETNOD: Failed at node REACH (2.298), Item 0 of 1
>>> SETNOD>
>>>
>>> The high level code to do the entry is in JNTMAN. It loops through
>>> the table passed to it via .NDINT, calling a lower level routine
>>> called SCTAND in SCLINK. An error here is passed up to JNTMAN, but
>>> it is not passed back to the user. There are some other problems in
>>> SCLINK pertaining to negative return values, so some minor work is
>>> necessary there, also.
>>>
>>> I'll make some changes to these two modules, generate a new monitor
>>> for VENTI2 and see what happens in a few days.
>>>
>>> Right now, if any Tops-20 using is using SETNOD to update DECnet
>>> tables, this appears to fail. If anybody else is seeing it or can
>>> reproduce it, I'd like to hear about it.
>>>
>>> On 5/4/21 11:15 AM, Thomas DeBellis wrote:
>>>
>>> Has anybody ever seen SETNOD fail to insert the entire node
>>> list? I
>>> just did.
>>>
>>> Shortly after I put my 20's up on HECnet, I wrote a reoccurring
>>> batch job that fires once a week on Sundays to pull the latest node
>>> list (T20.FIX) from MIM::. I use the highly venerable FILCOM
>>> program to do a difference of it with the previous week's list. I
>>> don't do anything in particular with the output except save it in
>>> case I feel like looking at it for some reason.
>>>
>>> The batch job always inserts the entire list, rewriting whatever
>>> might be in the monitor's data base. I have always been
>>> unsatisfied
>>> with doing things that way because it seemed to me to be
>>> inefficient
>>> as the node list grew. The HECnet node list count was 716 on
>>> 9-Jun-19 and it's now up to 884 as of the latest version that I've
>>> pulled, 30-Apr-21. The other problem is the microscopic
>>> possibility
>>> that a node is in Tops-20's monitor database (a hash table) that
>>> isn't in the HECnet node list.
>>>
>>> Nodes can get removed, although I think that infrequent. Nodes
>>> could get inserted outside of the batch job, but I think that most
>>> unlikely in my situation. Nodes can get renamed, as evidenced by
>>> 2.299 below, which went from THEPIT to THEARK. None of this should
>>> or has broken anything.
>>>
>>> However, it's been in the back of my mind to do two enhancements,
>>> one to Tops-20 and one to SETNOD. The NODE% JSYS should have an
>>> additional feature to return the current monitor data base. The
>>> SETNOD program should be enhanced to take that to compute the set
>>> difference with the new list. This would show additions, renames
>>> and deletions. That would bring the update operation down from
>>> some
>>> hundred items to less than ten, on average. This would obviously
>>> make more of a difference on huge DECnet's in the tens of thousands
>>> of nodes. Another NODE% feature should probably be to whack the
>>> entire monitor database except for the local node, which would be
>>> useful for trouble shooting.
>>>
>>> Last Sunday, the batch job failed with the following error:
>>>
>>> 18:33:40 USER SETNOD>*TAKE SYSTEM:NODE-DATA.TXT.0
>>> 18:33:40 USER
>>> 18:33:40 USER [Fork SETNOD opening <SYSTEM>NODE-DATA.TXT.1 for
>>> reading]
>>> 18:33:41 USER SETNOD>*SAVE
>>> 18:33:41 USER
>>> 18:33:41 USER [Fork SETNOD opening <SYSTEM>NODE-DATA.BIN.74 for
>>> reading, writing]
>>> 18:33:41 USER SETNOD>*INSERT
>>> 18:33:41 USER
>>> 18:33:41 USER *?SETNOD: Failed at node REACH*
>>> 18:33:41 USER SETNOD>
>>>
>>> I had a look at the SETNOD source and the HECnet node list and have
>>> discovered and concluded a few things. First, there doesn't
>>> seem to
>>> be anything syntactically wrong with REACH::'s definition: "set nod
>>> 2.298 name REACH". Second, there don't appear to be any semantic
>>> issues. 2.298 wasn't in use and it shouldn't matter if it was.
>>>
>>> In the case of INSERT, there are two kinds of errors from NODE%, a
>>> general failure of the JSYS and an incomplete insertion. The error
>>> is from the second case. Unfortunately, SETNOD isn't reporting
>>> enough information about the error, so I have to make some changes
>>> there. It's also possible that SETNOD is building an inconsistent
>>> database for the monitor to swallow; at least the LIST command is
>>> giving me some odd results, viz:
>>>
>>> SETNOD>list arEA 2
>>>
>>> [AREA 2]
>>> A2RTR
>>>
>>> TOTAL NODES FOUND: 1
>>>
>>> SETNOD>
>>>
>>> That's clearly wrong, viz:
>>>
>>> !i dec
>>> Local DECNET node: VENTI2. Nodes reachable: 7.
>>> Accessible DECNET nodes are: A2RTR BOINGO LEGATO
>>> TOMMYT VENTI2 VENTI ZITI
>>>
>>> The Exec output should probably be changed to say, "Nodes reachable
>>> in local area" and "Online nodes in area are:"
>>>
>>> Anybody have any ideas? Hunches? Clues?
>>>
>>> File 1) OLDF:[4,120] created: 1241 15-Apr-21
>>> File 2) NEWF:[1,1] created: 0102 30-Apr-21
>>>
>>> 1)1 set nod 44.9 name OSMIUM
>>> ****
>>> 2)1 set nod 2.292 name OSIRIS
>>> 2) set nod 44.9 name OSMIUM
>>> **************
>>> 1)1 set nod 13.3 name RED
>>> ****
>>> 2)1 *set nod 2.298 name REACH *
>>> 2) set nod 13.3 name RED
>>> **************
>>> 1)1 set nod 2.298 name RSX11M
>>> 1) set nod 1.306 name RSX124
>>> ****
>>> 2)1 set nod 1.306 name RSX124
>>> **************
>>> 1)1 set nod 42.5 name SPARKY
>>> ****
>>> 2)1 set nod 2.291 name SPARK
>>> 2) set nod 42.5 name SPARKY
>>> **************
>>> 1)1 set nod 2.299 name THEPIT
>>> 1) set nod 35.70 name THOMAS
>>> ****
>>> 2)1 set nod 2.299 name THEARK
>>> 2) set nod 35.70 name THOMAS
>>> **************
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sonic.net/pipermail/hecnet-list/attachments/20210505/3d059986/attachment-0001.htm>
More information about the Hecnet-list
mailing list