[9front] Re: drawterm stuck

[9front] Re: drawterm stuck

[William Gunnells]

Amavect

oh my net audit show everything seems to be fine except
ether=3D525400123456 does not belong to any network interface
no fs=3D entry  (needed for pls boot)=20

auth server seems to be fine i think it reads
someone is listening on port 567

auth/asaudit shows 1 problem

BAD: key in keyfs does not match nvram
trying nvram key for 9front@glenda with factotum
GOOD: key in factotum matches nvram

Re: [9front] Re: drawterm stuck

[hiro]

should it be using dom=localhost or dom=cirno.9front ?
it seems contradicting to me.
can you try without the complication of having multiple IPs (you seem
to be trying to use both localhost and that 10/24 subnet at the same
time)

Re: [9front] Re: drawterm stuck

[Amavect]

> oh my net audit show everything seems to be fine except
> ether=3D525400123456 does not belong to any network interface
> no fs=3D entry  (needed for pls boot)=20
Refer to my example from my last message and modify to your own setup.
You have duplicate sys=cirno lines, the cpu= line is not a '=', and your 
dom should be 9front, not cirno.9front.
Your ipgw and dns is your router's ip address, which is probably 10.0.2.1
No fs is fine if you're not planning to need tls boot.

> auth server seems to be fine i think it reads
> someone is listening on port 567
That's good. The cpu server is trying to listen for connections.
Clearly not succeeding, though.

> auth/asaudit shows 1 problem
> 
> BAD: key in keyfs does not match nvram
> trying nvram key for 9front@glenda with factotum
> GOOD: key in factotum matches nvram
I'm guessing you didn't set up glenda's password with auth/changeuser
Or, if you did, the key didn't match. Run auth/wrkey again.
Refer to the fqa links that I had put in my last message.

Thanks,
Amavect

Re: [9front] Re: drawterm stuck

[hiro]

and perhaps important fact: our error messages suck sometimes :)

Re: [9front] Re: drawterm stuck

[hiro]

reading the code it seems the server doesn't send valid
proto@authdom
pairs.

Re: [9front] drawterm stuck

[William Gunnells]

Okay I got it to work in virtual box. I struggled with qemu and tap/bridge to no end. 

I remember that virtual box handled bridge a little differently. So I gave it a shot and boom up and running. 

I do have OpenBSD on my laptop perhaps I can qemu working with tap and bridge.  But honestly I’m not sure what I want to do
in the grand scheme of things. 

I think it would have worked without the bridge but the qemu string would have been long because I think I needed to open up a range of ports for ingress and egress.

Thanks for all the help

Does httpd work and is it fast I was thinking of running this in AWS on a custom AMI at some point

Just looking for something different and something that could potentially be fast. I plan on serving controversial pages.

Its just hard to find bench marks but more importantly I didn’t want this to be the usual nix distributions. 



> On Jul 13, 2020, at 5:25 PM, Amavect <amavect@gmail.com> wrote:
> 
>> oh my net audit show everything seems to be fine except
>> ether=3D525400123456 does not belong to any network interface
>> no fs=3D entry  (needed for pls boot)=20
> Refer to my example from my last message and modify to your own setup.
> You have duplicate sys=cirno lines, the cpu= line is not a '=', and your dom should be 9front, not cirno.9front.
> Your ipgw and dns is your router's ip address, which is probably 10.0.2.1
> No fs is fine if you're not planning to need tls boot.
> 
>> auth server seems to be fine i think it reads
>> someone is listening on port 567
> That's good. The cpu server is trying to listen for connections.
> Clearly not succeeding, though.
> 
>> auth/asaudit shows 1 problem
>> BAD: key in keyfs does not match nvram
>> trying nvram key for 9front@glenda with factotum
>> GOOD: key in factotum matches nvram
> I'm guessing you didn't set up glenda's password with auth/changeuser
> Or, if you did, the key didn't match. Run auth/wrkey again.
> Refer to the fqa links that I had put in my last message.
> 
> Thanks,
> Amavect

Re: [9front] drawterm stuck

[hiro]

if you don't know how to do networking with qemu some people like to
use the libvirt abstraction around it that tries to do most of these
things for you (more like vbox).

personally i think it's better to learn how to do the networking right
with bridges and IP forwarding instead of silly hacks like NAT or port
forwarding.
qemu doesn't really need to do much here, the bridging or IP routing
would need to happen on the outer OS.

httpd is slow, but for controversial pages that might be fine as
you'll have less visitors.
all that benchmarks would show is that httpd is slow, so i don't why
you want benchmarks.
i recommend rc-httpd, it's slower than httpd. it's really good.

On 7/15/20, William Gunnells <gunnells@gmail.com> wrote:
> Okay I got it to work in virtual box. I struggled with qemu and tap/bridge
> to no end.
>
> I remember that virtual box handled bridge a little differently. So I gave
> it a shot and boom up and running.
>
> I do have OpenBSD on my laptop perhaps I can qemu working with tap and
> bridge.  But honestly I’m not sure what I want to do
> in the grand scheme of things.
>
> I think it would have worked without the bridge but the qemu string would
> have been long because I think I needed to open up a range of ports for
> ingress and egress.
>
> Thanks for all the help
>
> Does httpd work and is it fast I was thinking of running this in AWS on a
> custom AMI at some point
>
> Just looking for something different and something that could potentially be
> fast. I plan on serving controversial pages.
>
> Its just hard to find bench marks but more importantly I didn’t want this to
> be the usual nix distributions.
>
>
>
>> On Jul 13, 2020, at 5:25 PM, Amavect <amavect@gmail.com> wrote:
>>
>>> oh my net audit show everything seems to be fine except
>>> ether=3D525400123456 does not belong to any network interface
>>> no fs=3D entry  (needed for pls boot)=20
>> Refer to my example from my last message and modify to your own setup.
>> You have duplicate sys=cirno lines, the cpu= line is not a '=', and your
>> dom should be 9front, not cirno.9front.
>> Your ipgw and dns is your router's ip address, which is probably 10.0.2.1
>> No fs is fine if you're not planning to need tls boot.
>>
>>> auth server seems to be fine i think it reads
>>> someone is listening on port 567
>> That's good. The cpu server is trying to listen for connections.
>> Clearly not succeeding, though.
>>
>>> auth/asaudit shows 1 problem
>>> BAD: key in keyfs does not match nvram
>>> trying nvram key for 9front@glenda with factotum
>>> GOOD: key in factotum matches nvram
>> I'm guessing you didn't set up glenda's password with auth/changeuser
>> Or, if you did, the key didn't match. Run auth/wrkey again.
>> Refer to the fqa links that I had put in my last message.
>>
>> Thanks,
>> Amavect
>
>

Re: [9front] drawterm stuck

[Steve Simon]

re: httpd is slow

i think this is a little unfair.

i’m sure httpd would not stand up well to heavy loads - it wasn’t been designed for super high performance.

However i would expect it to keep up fairly well with light loads, after all it doesn't try to do too much.

imho file server performance is likely to effect it more.

-Steve


> On 15 Jul 2020, at 10:16 am, hiro <23hiro@gmail.com> wrote:
> 
> if you don't know how to do networking with qemu some people like to
> use the libvirt abstraction around it that tries to do most of these
> things for you (more like vbox).
> 
> personally i think it's better to learn how to do the networking right
> with bridges and IP forwarding instead of silly hacks like NAT or port
> forwarding.
> qemu doesn't really need to do much here, the bridging or IP routing
> would need to happen on the outer OS.
> 
> httpd is slow, but for controversial pages that might be fine as
> you'll have less visitors.
> all that benchmarks would show is that httpd is slow, so i don't why
> you want benchmarks.
> i recommend rc-httpd, it's slower than httpd. it's really good.
> 
>> On 7/15/20, William Gunnells <gunnells@gmail.com> wrote:
>> Okay I got it to work in virtual box. I struggled with qemu and tap/bridge
>> to no end.
>> 
>> I remember that virtual box handled bridge a little differently. So I gave
>> it a shot and boom up and running.
>> 
>> I do have OpenBSD on my laptop perhaps I can qemu working with tap and
>> bridge.  But honestly I’m not sure what I want to do
>> in the grand scheme of things.
>> 
>> I think it would have worked without the bridge but the qemu string would
>> have been long because I think I needed to open up a range of ports for
>> ingress and egress.
>> 
>> Thanks for all the help
>> 
>> Does httpd work and is it fast I was thinking of running this in AWS on a
>> custom AMI at some point
>> 
>> Just looking for something different and something that could potentially be
>> fast. I plan on serving controversial pages.
>> 
>> Its just hard to find bench marks but more importantly I didn’t want this to
>> be the usual nix distributions.
>> 
>> 
>> 
>>>> On Jul 13, 2020, at 5:25 PM, Amavect <amavect@gmail.com> wrote:
>>> 
>>>> oh my net audit show everything seems to be fine except
>>>> ether=3D525400123456 does not belong to any network interface
>>>> no fs=3D entry  (needed for pls boot)=20
>>> Refer to my example from my last message and modify to your own setup.
>>> You have duplicate sys=cirno lines, the cpu= line is not a '=', and your
>>> dom should be 9front, not cirno.9front.
>>> Your ipgw and dns is your router's ip address, which is probably 10.0.2.1
>>> No fs is fine if you're not planning to need tls boot.
>>> 
>>>> auth server seems to be fine i think it reads
>>>> someone is listening on port 567
>>> That's good. The cpu server is trying to listen for connections.
>>> Clearly not succeeding, though.
>>> 
>>>> auth/asaudit shows 1 problem
>>>> BAD: key in keyfs does not match nvram
>>>> trying nvram key for 9front@glenda with factotum
>>>> GOOD: key in factotum matches nvram
>>> I'm guessing you didn't set up glenda's password with auth/changeuser
>>> Or, if you did, the key didn't match. Run auth/wrkey again.
>>> Refer to the fqa links that I had put in my last message.
>>> 
>>> Thanks,
>>> Amavect
>> 
>> 

Re: [9front] drawterm stuck

[hiro]

to call something slow in such vague way is unscientific at best, and
you seem agree it's slow with heavy loads.

i claim that it's good enough, and i feel like you're saying the exact
same thing, what is unfair about what i said, or where do we even
disagree?

Re: [9front] drawterm stuck

[William Gunnells]

For such genius and beautiful OS why a slow web server. It kind of doesn’t make sense. Let me know if I’m missing something. Even if its file based shouldn’t there be some level of speed. Like hash tables seem to be extremely fast for lookup. Is it because they wrote httpd in one or two days?

Would it be better to have httpd dedicated to its own CPU and serve other files somewhere else. Serve files on a beefy server. Or am I thinking about that wrong. Should httpd be together? How does that work?

I realize I have a lot to re-learn about this. But I’m very interested. 

Surely this should be faster than some python based Django server running gunicorn?


> On Jul 15, 2020, at 6:43 AM, hiro <23hiro@gmail.com> wrote:
> 
> to call something slow in such vague way is unscientific at best, and
> you seem agree it's slow with heavy loads.
> 
> i claim that it's good enough, and i feel like you're saying the exact
> same thing, what is unfair about what i said, or where do we even
> disagree?

Re: [9front] drawterm stuck

[Kurt H Maier]

On Wed, Jul 15, 2020 at 10:02:30AM -0700, William Gunnells wrote:
> For such genius and beautiful OS why a slow web server. It kind of doesn’t make sense. Let me know if I’m missing something. Even if its file based shouldn’t there be some level of speed. Like hash tables seem to be extremely fast for lookup. Is it because they wrote httpd in one or two days?
> 
> Would it be better to have httpd dedicated to its own CPU and serve other files somewhere else. Serve files on a beefy server. Or am I thinking about that wrong. Should httpd be together? How does that work?
> 
> I realize I have a lot to re-learn about this. But I’m very interested. 
> 
> Surely this should be faster than some python based Django server running gunicorn?

nobody gives a shit about optimizing web servers

Re: [9front] drawterm stuck

[hiro]

if you have 64 cores, just run httpd 64 times and roundrobin each
request to one of them...

Re: [9front] drawterm stuck

[ori]

> For such genius and beautiful OS why a slow web server. It kind of doesn’t
> make sense. Let me know if I’m missing something. Even if its file based
> shouldn’t there be some level of speed. Like hash tables seem to be
> extremely fast for lookup. Is it because they wrote httpd in one or two days?

Because it doesn't need to be fast -- at least for what the people using it
seem to need-- and given a choice between "simple, fast enough to do the job",
and "a ton of complexity spent on optimization", the former is almost always
the preference of the plan 9 community.
 
> Would it be better to have httpd dedicated to its own CPU and serve other files
> somewhere else. Serve files on a beefy server. Or am I thinking about that wrong.
> Should httpd be together? How does that work?

Latency hurts 9p. Probably doesn't matter if the file server is nearby, but
you don't want to put it across the internt.

> I realize I have a lot to re-learn about this. But I’m very interested. 
> 
> Surely this should be faster than some python based Django server running gunicorn?

Probably. If you care, should be easy enough to point one of the
many benchamrks out there at it, and see.

Re: [9front] drawterm stuck

[Ethan Gardener]

On Wed, Jul 15, 2020, at 6:02 PM, William Gunnells wrote:
> For such genius and beautiful OS why a slow web server. It kind of 
> doesn’t make sense.

Because geniuses traditionally hate the horrible pile of amateur mistakes, stupid difficulties, and sheer brokenness that is web technology. >;) I wrote rc-httpd in an obsessional phase 11 years ago, then stopped paying attention to it. Remarkably for one of my projects, it turned out well (for a shell-script web server). A couple of years later, sl decided to put cat-v.org onto Plan 9 with rc-httpd, and it promptly got hammered by 4channers. 9front's kernel got some fixes to handle the heavy shell-script load. I think listen(8) got the maxprocs option at that time too. cat-v.org is still on rc-httpd (and werc; a bigger set of shell scripts) so i guess it can't have been too bad, although i don't think the heavy load lasted long.

A point to note when you're configuring your virtual machines: rc-httpd and werc benefit greatly from SMP. Shell pipelines automatically take advantage of it.

Re: [9front] drawterm stuck

[William Gunnells]

I don’t get a prompt or anything. I can however type on the screen. But commands don’t work. I would send a screenshot but I think that is filtered by the list server or something. I know that I make a connection but nothing is in the /sys/log/auth except for when I added glenda a few days ago using auth/changeuser

when I do an fshalt -r it gives an error in drawterm window
./drawterm: bad p9any domain
panic: ending

This is just a base install using 9front documentation online. I guess I did the following:
1. run the qemu install -> this seemed to be straightforward
2. initially i didn’t mess with compiler because I thought the default kernel would be fine. I did give up and compile this thinking maybe I’m missing something. 
3. change /n/9fat/plan9.ini -> service=cpu
4. change /lib/ndb/local -> the details of that would be in a previous thread. But in short its just the ip’s I found from ndb/query 
which were 10.0.2.0 etc…
5. I was not sure if I need to change authdom. I left it 9front I guess I could change it to whatever I wanted. 
6. I left everything cirno
7. as soon as I reboot i believe I had bad nvram. i did authid glenda and authdom 9front sec key some_password and password some password
Really everything is the same password because I didn’t understand that part
i made sure to have bootargs=local!/dev/sdXX/fscache -a tcp!*!564

Honestly that should be it
I checked netstat -na | grep -i listen | grep tcp which shows:
127.0.0.1.9567
127.0.0.1.17010

and i run drawterm -a tcp\!127.0.0.1\!9567 -h tcp\!127.0.0.1\!9567 
It seems I needed to put escape the ! and I was not sure if I needed a different port for -a 17010 neither of them seem to work
no errors. 

I’m trying to think if I miss anything. firewall is off. Drawterm compiled fine on mac. 

I didn’t think i needed to mess with any other filer like /rc/bin/cpurc ? 
Back in the old days you had to mess with a lot of files just to get things working. Honestly I forgot how to use this. 

here is something I just noticed in /sys/log/listen

started on tcp
giving up on tcp!*!567: (**

what’s up with that and why?




> On Jul 13, 2020, at 3:24 PM, Kurt H Maier <khm@sciops.net> wrote:
> 
> On Sun, Jul 12, 2020 at 07:48:42PM -0700, William Gunnells wrote:
>> 
>> I get no errors and no prompt. 
> 
> What prompt are you expecting to get?
> 
> khm

Re: [9front] drawterm stuck

[Kurt H Maier]

On Sun, Jul 12, 2020 at 07:48:42PM -0700, William Gunnells wrote:
> 
> I get no errors and no prompt. 

What prompt are you expecting to get?

khm