hey, i have a bit of a j2ee newb question, i'm trying to follow http://weblogs.java.net/blog/rampsarathy/archive/2007/03/glassfish_v2_an.html to get GF to to talk with activemq

i'm a little confused about jndi though

he links to this http://weblogs.java.net/blog/rampsarathy/archive/Main.java, but I can't figure out how that class gets invoked by Glassfish

slyphon: i dont see why that class would get invoked by glassfish

hmm, ok, i'm totally out of my element here

is that used to bind a resource in jndi?

and that toally looks like an assinine way to use glassfish and activemq

slyphon laughs

isnt there a jca way of doing it ?

java connector api?

yes

slyphon looks for the link

http://activemq.apache.org/sjsas-with-genericjmsra.html

yes, there ya go

i'm afraid i'm about as lost with this, though

although i got the first step done

generally, what i'm trying to accomplish is that my apps would be able to get a connection to activemq via JNDI

am i even on the right track?

and that configuration would pretty much live on glassfish

well, how do your apps normaly get a connection to activemq ?

(we're a ruby shop and moving to jruby, so it'd be nice to be able to tell the developers, "To get a connection that will work in any environment, just get a connection using 'jms/TehQueuz'"

well

well, why are you using activemq specifically?

create an ActiveMQConnectionFactory, do createConnection, etc

it seemed to be the best choice, as a standalone thing

it was popular, recommended by someone whose opinion i trust, etc. etc.

is that the same with glassfish ?

same reason for using glassfish ?

yeah, pretty much, plus we're a solaris shop, and everyone in the jruby community seems to have very good opinions of it

i'm really a j2ee neophyte

it *looks* like this would make an ideal deployment environment for our app

if there's no coupling of the ruby stuff to activemq... i'd recommend dumping activemq and just using openmq

well

there's no coupling, but why move to openmq?

less fiddling and more reliability

orly?

yhrly

:D

plus it's bundled in glassfish so... even less fiddling

ah, that's good

just sets up your destinations and go

how hard is it to set up failover/clustering/HA/replication/etc

you know, the Things That Let You Avoid SPOFs

slyphon: not too hard... but the difficulty in setting is pretty linear the complexity of your requirements

er, difficulty in setting it up

we have a pretty simple use-case, i think

i mean

if you just need HA JMS then it's pretty simple

yeah, that's us

it's all built into glassfish

oh nice

i have activemq set up now with two master/slave pairs replicating between both

but yeah, less configuration == less configuration

and interestingly enough im hearing more and more references of people dumping activemq for openmq for reliability and stability

oh, well

i havent touched activemq for like 3 years through myself

you had me at "reliablity...built in"

it seems that they've made quite a lot of progress, then again, it's from The People Who Brought You Apache Web Server

heh ive read through this blog like 6 times now and i still dont understand the steps required

oh thank god

i'm not a total moron

this ramesh guy is an openmq guy though... wonder why he's blogging about activemq and websphere mq

probably since he works on the generic jca adapter stuff

yeah, a bit odd

heh but if you look on glassfish/openmq forums... he helps everyone :)

unless im thinking of a different ramesh

slyphon: so this "replication" stuff you setup with activemq... that is to support HA?

slyphon: in an active/passive way ?

yeah

failover HA strategy

hehe, ActivePassiveMQ

you can define it in your connection string

:D

and it'll replicate that between whatever-nodes-in-each-cluster-is-up-right-then

s/is/are/

it duplicates the messages to all the nodes specified in your connection string ?

shouldnt it just send to one and if that fails, send to the next one ?

the activemq nodes know about each other and replicate the data to each other? or use a shared message store?

sorry, brb

[pwgr]: well, you set up your servers in failover pairs (they use flock(2) afaict) and a shared message store (on NFS or a SAN), so if one process dies the other process will acquire the lock and already have all the messages ready to serve out

[pwgr]: the way i configured it was having two pairs like that A-A' and B-B'

tehn you set up a transport connection between the two of them (in the config file)

so B-B' <-> A-A'

over NFS? that's horrible

ehhh, it's not as bad as you might think, if you have a reliable server

well, i know exactly how bad it is. hehe

plus, we have NFS pretty much everywhere, so it was a pretty easy thing to roll out

heh :)

yeah

NFS F**king Sucks

[pwgr] has too much experience with storage internals

:)

but SAN isnt that bad if you have it

yeah, unfortunately we don't

it uses a raw device or something ?

not really sure

i think they meant "shared filesystem"

with locking

or do they assume SAN == "expensive clustered filesystem on a SAN"

it wasn't an option for us, so we couldn't use that

yeah

i think that's probably more accurate

i *seriously* doubt they're doing anything at the block level

really? that seems the best way to go

well

i mean

"best" in one way, sure

but, you know, "windows"

w/ NFS you have the full IP stack to deal with, and then local platform locking semantics and whatever bugs and interoperability snafus lies with in that

absolutely

how does openmq deal with it?

well actually if you're running on solaris its not the end of the world

:)

yeah, solaris has been very good to us

but would be much better to be able to setup an iscsi device and go raw block dev access

except for a 10u4/u5 NFSv4 lib conflict

*that* was *hella* fun to try and figure out

"WTF is 'ESTALE'?!"

openmq uses a clustered JDBC datasource for clustering

hmmm

can you use derby?

hehe nah derby's not clusterable

*damn*

you'd use something like oracle rac

slyphon was hoping they'd added that

fuck

but actually it's probably preferred to just use HADB

dude, if we had a budget...

HADB is no cost

HADB?

unfortunately we don't have an RDBMS solution at my place

we have MySQL

which is good...you know...if you don't try to do anything with it that you couldn't do with sed/grep/awk

yeah HADB is sun's HA database

it's not opensource but it's free

ahhhh

is it part of cluster?

slyphon phears sun cluster

ohhhh

riiiiight

so *this* is what the "HA Session Store" is

nah it's not part of sun cluster

though sun cluster might have agents for it

slyphon nods

i tried to read through sun cluster's documentation once

after i got a nosebleed and passed out, i decided that was enough

so is HADB distributed with GF as well? oh, and what's the difference between GF and the Sun Application Server that's bundled with their directory server?

HADB isnt bundled with GF or SJSAS, it's a separate download, but usually linked with the SJSAS downloads

slyphon nods

actually they may have started bundling it together but i havent looked in awhile

slyphon nods

what's the diff between GF and SJSAS?

SJSAS is just a commercial distribution of GF

"pizzaz"?

ah

they are very much the same thing

oh

is it at all like solaris v. opensolaris?

or not even?

it's not even that big of a different

oh, cool

well, that'll be easy to sell to the CTO :)

but, sorta the same idea business model wise

slyphon nods

to be honest, it'd be easier to deploy something that had native packages anyway

i wouldnt run anything critical on opensolaris :)

no, we don't

10u5, baby

zfs + zones ftw

well, i just like having patches available

this is true

oh, they renamed it

i care more about real patches and timely fixes for stuff more than bleeding edge features like zfs roots and stuff

it's now the Sun GlassFish Enterprise Server

yeah, totally

although, having the virtualized network stuff they've been talking about for zones would be sweet

yes it would... very much looking forward to that

have been waiting for that to come to solaris for years actually

yeah, we had a hell of an adventure with multi-homed hosts and zones

resource managment and stuff has been wonderful at the cpu and memory level in solaris... but still very much lacking in the networking realm

unfortunately, my team has very talented sysadmins and systems developers, but networking ninjas we ain't

slyphon nods

i was really suprised at how solaris didn't even really have a decent firewall (aside from ipfilter or what-have-you)

yeah... they got the ipf stuff down... the bsd style firewall stuff... but ipqos is still lacking

slyphon nods

but for me ive always had better luck with solaris and the primitive ipf than linux's equivalents

oh yeah

i mean

i had so many pobs with the likes of ipmasq and ipfwadm or whatever that was

i was really amazed at how much more solid solaris is

and i still have some problems with iptables

slyphon nods

and it's like all the fixes are "get this newer kernel version and/or apply these kernel patches"

i think pf is the only firewall config i've ever understood, and i'll be *damned* if i'm gonna run openbsd

that's not really an option for me

slyphon nods

which is why i'm not running opensolaris and friends :)

it's fun if you like bleeding a lot, and getting paged constantly

yeah IMO once things go beyond ipf... time to get a dedicated firewall

:)

although

we have two ASAs and they're a *bitch* to configure

getting NAT working on them was *not* fun

and my networks are never big enough to have dedicated network admins

slyphon nods

anytime i have anyone dedicated to networks... they dont have enough real work to do so they are artificially bloating requirements to make projects for themselves or increase complexity to keep themselves stimulated

yup

so, sysadmins that have social networks to draw from where we can have real network guys when we need them and have them out of the way when network is fine...

heh

those social bonds get guys who want to do a good job for the sake of their relationships with their friends... it just produces good results :)

and that's why "consulting" was invented

yeah, totally

whereass getting a consultant from a local VAR usually, in my experience, ends up in consultant doing halfassed job that requirements more maintenance than is necessary

well, yeah

i mean

i like the guys who come in, do the job, xfer the knowledge necessary to the admin staff, hand me an invoice and get the heck out

how else are they gonna guarantee you're gonna call them back?

exactly

that's just bad business, in the end

yeah looks like they are pushing the Glassfish Enterprise Server name now

i guess it's popular enough as a brand

yeah i was gonna say they surprisingly got the glassfish brand rolling in the past two years or so

so why not take advantage of it

indeed

it's recognizable, that's for sure

sjsas_ee-9_1_01-solaris-i586.bin Sun Java System Application Server 9.1 Update 1 with HADB 220.99 MB

i guess they are bundling it altogether now

how convenient

slyphon is downloading it right now

HADB isn't hard to set up?

nah, not if you know how to read the documentation

:)

yeah, that's another thing that i was really pleased with when we moved to solaris

they actually *have* documentation

but there's more to it than simply running an ant script

heh

you have to do stuff like setup networking, clock syncing, some IPC stuff

woah

that's heavy-duty

hmm, do you think you could use a LAG instead of IPMP?

does LAG do failover ?

or survive partial link failure?

well, yeah, it's 2 interfaces 1 ip

either goes, you're still up

but it survives if one interface goes tits up and is left with less than initially setup with ?

yeah

the only disadvantage is that you have to have both wires plugged into the same switch

i'd bet it'd work

you just have to choose how much redundancy you actually want

how many points of failure you want to account for

i mean, the point is "This Physical Link Will Not Go Down"

slyphon nods

well, initially, i ain't gonna bother with that ;)

ugh

i wonder if i can get away with not editing /etc/system and rebooting :/

oh, editing /etc/system fo the semaphore stuff... yes, you can

use project stuff

ah

how about zone config?

i think they let you set some of the shmem stuff in your config

hrm, i dunno about that

eh, i'm probably not gonna set this up tonight anyway :)

i'll need to digest it a bit

i know you can set memory caps and cpu scheduling stuff, but i dunno about IPC stuff

slyphon nods

i've never had to think about IPC so i'm not sure myself

hmm, im thinking you can set zone attributes in resource configs

The zone.max-msg-ids, zone.max-sem-ids, zone.max-shm-ids, and zone.max-shm-memory

resource controls are used to limit System V resources used by all processes within a zone.

hehe yep that's them right there

yay!

ok, there's an HA admin guide in the SJSAS docset

it covers JMS

*sweet*

i was reading this: http://docs.sun.com/app/docs/doc/819-3679/abdca?a=view

yep that's the HA admin guide

kewl, man, they must have an *army* of people writing this stuff

slyphon blinks

[root@overkill ~]# xclock

hehe it's probably like 3 dedicated people :)

Segmentation Fault (core dumped)

hahaha

now that ^^ is something you don't expect to see

[admin@solaris1 ~]$ xclock

Error: Can't open display:

that's what i expect :)

hahaha

WHOAH

?!

i logged in with ssh -X

and accidentally typed xlock instead of xclock

!

heh locked my local X server

hahahaha

luckily it said login name was the admin user

i would have had no idea how to unlock it otherwise

that would have *sucked*

and im multitasking between about 14 different things here

:)

would suck ot have to kill the Xserver and try to figure out everything i had going again

oh yeah

gnu screen, ftw

yeah i have vncs to a bunch of servers and remote vmware connections and stuff

ahhh

yeah

the guy that figures out 'screen' for X is gonna be famous

im testing pca on solaris... first time ive tried it

oh, i really like it

and i'm a total n00b when it comes to patching solairs

which may be a vote for or against it in your book :)

if it works for me i'll like it

otherwise I wont :)

but i was able to patch our systems to roll out sun DS without really having to know what i was doing

it's the closest to apt-get i've found on solaris

smpatch treated me well for a few years but lately its really messin up on me

i liked it initially, but the problem was that i'd log in and there'd be like 100 ccctransport proceses running

sunDS ?

I use it but i dont use the solaris package versions... i install the standalone tarball

and dont bother with console integration and that nonsense

directory server

yeah... i'm not familiar with what ccctransport is

common container connection transport? :)

something like that

[pwgr] guesses wildly

it's like CCCP for unix

i just like doing the directory server stuff from the commandline

oh? i'm glad they have the whole webconsole thing

b/c LDAP and me? we ain't friends

we're cordial, but wary

heh yeah... i use the web admin thing inside glassfish for testing

that little struts based thing they have that comes with the directory server

yeah

they bundle a war but dont bundle a tomcat instance or anything :/

:P

so i just plop it into a glassfish domain and use it from there

hahahaha

nice

works as well as is needed

it's meant to integrate with webconsole (svcadm enable webconsole)

but it's a bitch to get it all working smooth

hehe yeah... me and webconsole aint friends

:)

thankfully someone else from work figured it out and documented it

that's one part of solaris i havent bothered to work through

the biggest problem we've had with it is that *nowhere* is it documented how to replace the SSL cert

the webconsole or the directory server ?

webconsole

oh

the DS we have working with our internal CA

but not the webconsole ?

yup

it's not a huge deal, it just means firefox bitches :)

i wouldnt know anything about it

yeah

i'd really only need it if i want to hand stuff off to some nonunix people

slyphon nods

i have some customers that run some linux stuff very blindly

hehe using webmin

:)

oy vey

webmin got me off the ground when i was starting out

i havent tried the webmin stuff in solaris too much... i think i set it up and was trying to do some sendmail config with it and it broke and i didnt bother after that

but figured webconsole would be better suites

suited

slyphon nods

deploying that war in GF sounds like a good workaround

actually i kinda liked the volume management stuff that was in webconsole

oh? i do all that zfs stuff by hand

for setting up concatened volumes for UFS filesystems and stuff

ohhhhh

yeah, that'd be useful

yeah i havent tried any of the zfs or containers stuff in webconsole

i just dont like how much memory it chews up :)

hahaha

and i dont like all the security vulnerabilitites and patches i see for it

well

i mean

don't expose that to the web, duh

;)

does the webconsole stuff work OK in firefox other than the SSL complaints?

oh, definately

they improved the interface in u5+patches

hehe man i just started using firefox3 and it seriously pitches a fit w/ non 3rd party authorized stuff

haha

yeah, at least they let you store the cert

which is an improvement

hmm i never store certs that arent validated by a known authority

i store self signed certs but other than that... like certs that are generated for testing purposes, i never store those

oh?

oh, yeah

im too afraid of forgetting about them or something

our shop is sufficiently small where we don't really have to worry too much about that

we're not hosting anybody else

finally! done patching

pca definately takes longer than smpatch :/

hm

well, it *is* like a 5000 line perl script

[pwgr]: oh, here's a n00b question for you, can you control what uid/gid a given app runs as in GF?

nah everything runs as the uid/gid running the jvm

soh!

sadlkfjas

doh!

i just standardize on a single app server user, install glassfish as that user, and everything runs as that user

there's no concept of setuid in jvm land without resorting to skanky trickery

yeah, we've pretty much got that

slyphon nods

thank goodness solaris allows you to grant privs to nonroot users to run stuff on ports less than 1024

yeah, seriously

the RBAC stuff is impressive

cant believe linux doesnt support that

linux is the future!

i think you really have to *pay* people to develop shit like that

haha

and it always will be

root

ewps

got too much stuff going on here, grabbed the wrong mouse

hah

eeeeeek!

[pwgr]: ok, bedtime for me, good talkin' with ya

thanks for the help

yep same, gnite

lata

hi all - I'm trying to setup Javamail, and it's insisting on a username for STMP, however... it's a internal relay, so no auth is required

pookey: maybe this question is better asked in #java

a username of '-' makes glassfish accept it, and my mail server is relaying just fine - seems a bit silly that the glassfish UI makes it required, perhaps I should open a bug report

if I'm monitoring a connection pool, do I enable monitoring in the DAS 'server-config' or the cluster 'roller-cluster-config'?

Anyone knows how to a hold of a datasource with JNDI from within a custom realm implementation?

why would a values retrieved through a session bean and Entity classes doesn't match the data in the database table?

this only happens when I update the records through my RDMS client

my guess is that i'm seeing caching in action, if this is the case how do you stop it?

yes, its a caching problem,

redrebel: I think there's a flush() method on the entity manager.

if i wait long enough, the values do change

if something else is changing the data without letting the app server know about it... that will happen

that is what is happening

em.flush() by itself is not helping

if the session+entity setup in the app do not have exclusive access to the db... you'll have to either make everything else notify the appserver of the data changes, which is not trivial... or yu'll have to modify the querying of the session beans to refresh the data

nah flush isnt what you need... more like refresh()

redrebel: refresh() was what I was referring to. Just forgot the name :)

flush() is the other way.

IMO, refresh() from the entity manager isnt even that good of a way to do it

better to tell the jpa provider to refresh during the query process

done through query hints if you have experience with those

if you're using glassfish w/ the bundled TLE provider you can do something like setHint("toplink.refresh", true) or something

i guess itd be setHint("toplink.refresh","true") since it wants a string as the 2nd param

i think i'm just gonna stick to managing the db through my session beans

nikolaj_a: yeah flush is the opposite direction :)

flush is an awesome name for an api function!

[pwgr]: Yeah, I have fields named 'toilet' all the time :P

A bit like calling your exception variables for 'up';

And now I'n beginning to end my sentences with ; instead of. which means it's time to go home :P

nikolaj_a: haha yeah ive written a lot of buffering and caching code containing statements like "if (dirty) flush();