Ad Widget

**richlv** · 30-10-2010, 12:00

thanks for the feedback. just some quick notes.

Originally posted by aprudnev

Now, what is weak.
1) Virtual Appliance - good idea, bad implementation.

from what you have mentioned, you had the very first version of virtual appliance for zabbix. ever

as such, it indeed had some rough edges (most notably, dom package missing). appliance for zabbix 1.8.3 fixed these issues and one for 1.8.4 is expected to improve further

Originally posted by aprudnev

1.1) must be stored as ovf appliance, not as VM disk images.

it's important to note that zabbix virtual appliance has been prepared using suse studio - a quite nice solution that allows to create appliance very, very easily. so format of the images depends on what is available in suse studio.

aaaand... the great news is that suse studio recently added support for ovf, so expect zabbix appliance for version 1.8.4 to be provided in ovf format as well

Originally posted by aprudnev

1.2) must be well configured. I improved it's speed 3x by recreating database and fixed import/export by adding packages

mmm. would you mind sharing the secret behind 3x performance improvement ? that would be surely great to implement upstream as well.
broken import/export because of the missing dom package was discovered right after the release and fixed promptly. some sort of a lesson to learn from the very first version

Originally posted by aprudnev

(in addition I added full X11 stack because this is my requirement - any Linux must have X11 installed).

now that's different from some other view - that server solutions should never include a gui. additionally, it helps to keep the download size down, so that's not expected to change.

Originally posted by aprudnev

1.3) I love OpenSuSe but maybe using CentOs 5.4 is more practical for the VA purpose. Format must be, AGAIN, OVF (it is zipped and use open standard and you can install VM by import from http directly).

as noted, it's created using suse studio, so distribution is not expected to change. not seeing any arguments why centos would be more practical either

Originally posted by aprudnev

1.4) More and more systems became virtual today; zabbix can be used as virtual system in production installations. Maybe, it should be physical starting with 500 - 1000 monitored hosts, but it can definitely monitor 100 - 400 hosts running as VM (with a lot of items, not empty lists!).

not fully sure what's weak here

Originally posted by aprudnev

2) Configuration.
2.1) Application concept is not well developed. It can be much better. How can I create a group of items and triggers for the file system, and then replicate them for '/', '/usr', '/var' (for example). The same about processes and many other things. It all require manual replication.

great news - this specific example of item/trigger groups per filesystem should be much more automated in zabbix 2.0 with low level discovery.

Originally posted by aprudnev

2.2) Item copy should copy triggers as well.
2.3) Item cloning should clone related triggers as well (at least allow to do it).

debatable/non-trivial. once you consider the fact that triggers in zabbix can reference one item multiple times, different items - and even on different hosts - copying and cloning items poses a quite hard question what to do with triggers. it is quite likely that some simplified solution could be devised (only modify the item key of the copied/cloned item, for example), but it's something that should be seriously considered first

Originally posted by aprudnev

2.4) Full template clone should not link new template to the same hosts as old one (it really create a huge disaster sometimes).

i fully agree and it has been fixed in 1.8.3 - https://support.zabbix.com/browse/ZBX-2639

Originally posted by aprudnev

2.5) Item data type should be known better - it is now a good adventure to create item for disk IO monitoring, then guess, are data integer or float...

how would you propose to communicate that ? with zabbix agent built-in items, you can do zabbix_agentd -p and see item types, but it's not really feasible with any other types where server/frontend would have no idea what data will be returned

Originally posted by aprudnev

2.6) Can use $1 in item name; but why can't use $1 in trigger name?? It creates a lot of fun.

what would you expect it to resolve to, especially with more complex trigger expressions ? https://support.zabbix.com/browse/ZBX-3147 might be relevant.

also, what's the fun exactly ?

Originally posted by aprudnev

2.7) No way to find all triggers with the same or similar name, except some tricks in the search filters.

trigger filter in configuration has been discussed before and in general introducing one is considered a good idea

Originally posted by aprudnev

2.8) Trigger dependencies - can't set up dependency for the group.

dependency management is indeed a bit limited. dependencies are powerful, but managing them is not easy. implementing it properly isn't easy as well...

Originally posted by aprudnev

2.9) Template usage - I can change trigger severity when trigger is templated; can I change trigger threshold number? ideally I should specify: Application: FileSystem(/usr,Warning=15,Alert=1) and it should generate items for /usr with triggers from template with Warning trigger 15% and Alert trigger 1% (for example).

if you want to have custom thresholds in triggers per subtemplate/host, see user macros - there should be an example or two to get you going

Originally posted by aprudnev

2.10) No way to get list of objects (processes, file systems) from agent and then configure items based on this list.

as noted, this will be possible in 2.0 with low level discovery. actually, just released 1.9.0 (alpha) has a preview of this feature already.

Originally posted by aprudnev

2.11) Can I add last key value into the Subject in notification? (Not obvious how)

yes - see macros

Originally posted by aprudnev

2.12) Can I add a link to the chart for the trigger key value (extremely useful - when you got 'Cpu high , link to the cpu chart')

in frontend or in notifications ? although at this point it would be better to split such specific questions in their own threads to prevent birth of a monster thread

Originally posted by aprudnev

3) Presentation layer.
It don't make what it should. You can see events but can't see objects and their status until they really fail.

See what I mean. I would like to see something like snmpstatd ACTIVE or TOTAL or ERROR screens - they show all objects with (just all, or al active, or al with problems) and shows in a tool-bars current status such as traffic, errors, cpu and so on. See an example - I can see list of objects, their status, if something wrong - see the reason in snmpstat, but I can see only color and number in zabbix (can't click on group and expand it into the list of hosts with cpu, memory and disk op compact charts).

while presentation of current state, problems and values certainly can be improved, proposed view would be extremely limited, unless it is made very, very flexible. if it's not very flexible, i'd guess some 80-90% of zabbix users would find it too limited, because zabbix is not restricted to monitoring network traffic and cpu load - usually people monitor much more with it, and on such a view they would want to add all kinds of other information.

customisable, flexible view would be great to have, but it is also very hard to make right. and there hasn't been a huge interest in financing one

**aprudnev** · 01-11-2010, 11:13

Hmm, interesting to discover that I was unlucky to take VA few days before next had been released. And that 50% of observations are fixed already.

Few quick comments.

1) Virtual Appliance - ovfiormat is usually created by just exporting VM as virtual appliance; it is supported by Vmware ESXi (which is FREE anad which is THE ONLY production-grade free VMware - Vmware server is something better to forget about). So, while support of ovf format by OpenSuSe studio can be a great thing, I don't see why just don't create VM Linux, install and test everything and then export - and it can be ANY Linux.

CentOS vs OpenSuSe. OpenSuSe is a great system. But it is not production system - it has life span of 3 years, while CentOS has it 7 years. This is one of the reasons why OpenSuSe is rarely used as a server platform while CentOS is the most popular free Linux for the servers. It's not important for me - I can install from scratch and I know SuSe even better then CentOs/RedHat, but it can be some limitation for others. We used OpenSuse as a servers but switched to CentOS because of this life time issue (and then it is little more stable). Anyway, I agree that VA platform is not so important and OpenSuSe is a good choice.

X11 - I agree; while I require ALL servers to have GUI (it is disabled but ready for use, and it save a huge amount of time on troubleshooting and configuration), it is much easy to ADD x11 vs to remove, so VA should not have X11 except maybe xterm and 1 text editor (so if you slogin -X, you can use gui editor instead of ugly vi or emax).

Zabbix performance - VA had database with everything in a single file and with the weak mysql configuration (very small caches etc). Second is obvious (of course you don't want VA to use 4 GB of memory just for the demo), but first was a bug (my.cnf had table-per-file configuration but zabbix database was created without it. history tables partitioning can help as well. I don't know what exactly helped but I export zabbix database, imported it into newly configured mysql (VA was the same - 4 CPU. 3 GB memory, high priority, no CPU or memory reservations) and CPU usage dropped 3x - 4x. I did not investigate it, but of course VA should be configured more carefully.

2) Regarding trigger cloning and names. I mentioned it as an option just because it is not easy for multi item triggers. But it is easy for single irtem ones.

3) Regarding names:
- I have item which read disk space from disk /usr
- so I use $1 which means _first parameter in all item expressions_ and I see item as _Disk space on /usr
- Now I create trigger which uses this disk space. It has /usr as a first parameter, too
- but I can't create it as
Not enough space on disk $1

So I must write disk name manually every time, and it creates a hassle. It really works but takes first number in expression instead of first item argument (so it expands into 2 because expression is <free space> < 2, where free space is exact item for disk space - but it should expand into the same thing as in the item. I believe it is just a bug because it works but works the wrong way.

(A, exact example:

{ExM.Windows_general_tmpl:vfs.fs.size[c:,pfree].last(0)}<1

I write name as

Low free disk space on {HOSTNAME} volume C:

instead of

Low free disk space on {HOSTNAME} volume $1

because last one does not work as documented (first parameter in the item).

But I write an item name as

PFree disk space on $1

for
vfs.fs.size[c:,pfree]

so it works in the item

)

Originally posted by richlv

how would you propose to communicate that ? with zabbix agent built-in items, you can do zabbix_agentd -p and see item types, but it's not really feasible with any other types where server/frontend would have no idea what data will be returned

I agree that it is not so simple task. Ideally system should ask agent, get list of supported items with item data type, and then use it.

Maybe, just allow not fill this field in and then set data type on the first return from the agent? Or convert integer to floating?

Minor comment - system shows number of unsupported items. If I can open them all , with exact error message for each of them, it could simplify setting dramatically (I see 2 unsupported items now - without an idea what they are about - I can't click on this number and see them all).

Originally posted by richlv

what would you expect it to resolve to, especially with more complex trigger expressions ? https://support.zabbix.com/browse/ZBX-3147 might be relevant.

also, what's the fun exactly ?

Fun is a game _Guess why this item is reported as unsupported_ and _Find my 2 unsupported items over 1000 supported ones_.

Originally posted by richlv

in frontend or in notifications ? although at this point it would be better to split such specific questions in their own threads to prevent birth of a monster thread

I already open this thread in Russian branch of the forum, and consider writing it as a feature request with possible participation in development (as former snmpstat developer, so I invented this status bars many years ago and they appear to be extremely useful).

Proposed view is very flexible - I would like to
- describe status bar parameters -you can define up to 3 status bars, each of 1 or more /by multi color/ parameters;
- describe _SHOW_ rule - for example _CPU > 10% or network usage > 10% of disk io > 100KB/second.

And then it is EXTREMELY useful. As I explained, there is not other way to see WHAT IS GOING ON IN MY DATA CENTER_ when I have 100 - 1000 servers, and want to see a graphical presentation of their overall status. I show a few examples in the other thread. You can't set up thresholds and you can't use regular charts (because you can see maybe 10 - 15 objects while status bars allows to see 1000 ones, and filter allows to filter out all non-interesting ones). It works extremely well for the network and I don't see why it should not work for the servers (I show it to Proactive-Net architect 5 years ago and he was impressed how effective this is).

But let's keep this for the dedicated thread.

**zalex_ua** · 02-11-2010, 23:45

Originally posted by aprudnev

3) Regarding names:
- I have item which read disk space from disk /usr
- so I use $1 which means _first parameter in all item expressions_ and I see item as _Disk space on /usr
- Now I create trigger which uses this disk space. It has /usr as a first parameter, too
- but I can't create it as
Not enough space on disk $1

"item expressions" - is wrong, "item key" - is right !

So far, there is a problem - the macros are expanded in not all places of the interface.

Originally posted by aprudnev

It really works but takes first number in expression instead of first item argument (so it expands into 2 because expression is <free space> < 2, where free space is exact item for disk space - but it should expand into the same thing as in the item. I believe it is just a bug because it works but works the wrong way.

where did you get that it should work exactly?
The same is not yet documented! Richlv also talked about this!

Macros $1-$2 in trigger expressions (not the same as item key) are expanded not in the key values, it expanded into the right side comparison of each item in the trigger expression. Some examples 1 2

Originally posted by aprudnev

(A, exact example:
{ExM.Windows_general_tmpl:vfs.fs.size[c:,pfree].last(0)}<1

I write name as
Low free disk space on {HOSTNAME} volume C:

instead of
Low free disk space on {HOSTNAME} volume $1
because last one does not work as documented (first parameter in the item).

But I write an item name as
PFree disk space on $1
for
vfs.fs.size[c:,pfree]
so it works in the item

Try to use how zabbix_server expand this macros in notifications. And try to use macro {ITEM.NAME<1-9>} in trigger name. For notifications it should be work.

**aprudnev** · 03-11-2010, 00:17

Originally posted by zalex_ua

"item expressions" - is wrong, "item key" - is right !

So far, there is a problem - the macros are expanded in not all places of the interface.

where did you get that it should work exactly?

No matter how it is documented NOW, it is very logical - to have THE SAME meaning of $1 $2 etc in triggers and items including trigger names. N-th parameter of the item key. This way if we change an item (change disk name from / to /usr for example) it can automatically change trigger name exactly as ir changes item name (very convenient).

**zalex_ua** · 03-11-2010, 02:05

Originally posted by aprudnev

No matter how it is documented NOW.

Originally posted by aprudnev

it is very logical - to have THE SAME meaning of $1 $2 etc in triggers and items including trigger names. N-th parameter of the item key. This way if we change an item (change disk name from / to /usr for example) it can automatically change trigger name exactly as ir changes item name (very convenient).

$1 $2 in triggers names this is an entirely different section of analysts and is very useful. Live with it.

**aprudnev** · 08-11-2010, 23:50

Originally posted by zalex_ua

$1 $2 in triggers names this is an entirely different section of analysts and is very useful. Live with it.

I don't see any example, when it can be useful.

'Disk usage is more then $1 on disk <what here???>' - it is more or less useless because we clone triggers and items and change item keys but rarely change trigger thresholds.

And then how to put 'item key' into the trigger name? 'Low disk space on disk <what is here??>'

PS. Will post feature request for presentation layer in a few days.,

**zalex_ua** · 09-11-2010, 00:36

Originally posted by aprudnev

I don't see any example, when it can be useful.

'Disk usage is more then $1 on disk <what here???>' - it is more or less useless because we clone triggers and items and change item keys but rarely change trigger thresholds.

I expressed my own opinion, it may not coincide with yours.

Originally posted by aprudnev

And then how to put 'item key' into the trigger name? 'Low disk space on disk <what is here??>'

For example, use one usermacro {$YOUPARAM} as parameter for Item key and simultaneously simply in trigger name.

**aprudnev** · 09-11-2010, 00:44

Originally posted by zalex_ua

I expressed my own opinion, it may not coincide with yours.

For example, use one usermacro {$YOUPARAM} as parameter for Item key and simultaneously simply in trigger name.

I do not object your opinion - I just ask to show me any example, how you can use existing macro ($1 etc) in the triggers (they expand to the N-th right side number in the logical expressions) ? You said _it is very useful_ - maybe, but I can't find any example, when this can be useful, while my proposal ($1 is first key parameter and so on) can be used _out of the box_ (trigger name - low disk space on disk $1 - as an example).

User macro can't be used here because we can have 10 items with 10 different parameter keys (/, /usr, /var for disks; eth0, eth1, eth2 for NIC-s and so on) and we need to refer to this parameter in the related trigger names and trigger subjects... user macro can't be used because it is global for the whole server.

**zalex_ua** · 09-11-2010, 01:19

Originally posted by aprudnev

I do not object your opinion - I just ask to show me any example, how you can use existing macro ($1 etc) in the triggers (they expand to the N-th right side number in the logical expressions) ? You said _it is very useful_ - maybe, but I can't find any example,

Ok. I am preparing some articles on the wiki, I'll open the curtain

sorry, I do not know how this forum put a picture vertically, it's annoying me the most

Originally posted by aprudnev

User macro can't be used here because we can have 10 items with 10 different parameter keys (/, /usr, /var for disks; eth0, eth1, eth2 for NIC-s and so on) and we need to refer to this parameter in the related trigger names and trigger subjects... user macro can't be used because it is global for the whole server.

If you knew how the trigger expression is stored in the database, you'd understand why quite difficult to do what you want.
Think of compound expressions, where multiple Items. I will not argue - let the developers say.

Attached Files

**aprudnev** · 09-11-2010, 01:50

Your example shows few triggers for the same item. It can be useful but in rare cases (I have WARNING and ALERT for disk space, or you have few levels of alerts for the UPS). They are usually created once on the template and never cloned. You used user macro just as a replacement for the server (host) name.

It is all legitimate but do not solve a simple problem - we need to clone the whole groups of item + related triggers (network port, file system, process); when clone we change expression key (and nothing more) in most cases. We can then have automatically created item names which is good. But we can't clone trigger, change item key (or make automatically cloned triggers tied to the item) and see trigger name change when we change the key. And question is _HOW TO DO IT_. This can be very important when/if you add 'item groups' which group items and triggers with the same item function.

_HOW is it stored_ is another story; I (as a developer too) always understand _it is not safe to implement now_. But I don't see much use for existing syntax - you rarely clone triggers and then change threshold and need it in the message; you need _OBJECT NAME_ first of all (file system /ust; process 'zabbix-agent'; Voltage '+5V', and so on; and you need threshold only after it and it is not so important.

Let's compare possible error messages:

1 - The best message: File system /usr is full, usage is 92% threshold is 90%, here (url) is historical chart.

3 - Almost the same message: File system /usr is full, usage is 92% threshold is 90%,

4 - Not so good message: File system /usr is full, check it immediately.

5 - Almost useless message: File system is full (92%) (Which file system? Maybe it is /mount/cdrom file system which is always full?)

So, I need threshold last, and I need object name _event is about_ first. And I clone _objects_ much more often then _thresholds_ - I can have 100 processes and 20 file systems, and will have maybe 2 - 3 triggers per each.

Multi item triggers can exist but are relatively uncommon and so they are the very last concern (in the messaging) - and they often will use THE SAME parameter key (trigger: errors on eth0/in or eth0/out - still need interface name even if it maybe is constructed out of 2 items - but both are very very likely with the same parameter - interface name).

Originally posted by zalex_ua

Ok. I am preparing some articles on the wiki, I'll open the curtain

sorry, I do not know how this forum put a picture vertically, it's annoying me the most

If you knew how the trigger expression is stored in the database, you'd understand why quite difficult to do what you want.
Think of compound expressions, where multiple Items. I will not argue - let the developers say.

**zalex_ua** · 09-11-2010, 02:25

Originally posted by aprudnev

You used user macro just as a replacement for the server (host) name.

No, in my case the macro {$UPSNAME} is the first parameter for item key, which is used in UserParameter by zabbix_agentd, but it does not matter.
I repeat once again - I'm not going to argue with you. Please contact the developers with your ideas.

Originally posted by aprudnev

Let's compare possible error messages:

You do not accidentally forget about the macro {ITEM.NAME<1-9>} for notifications?
You can include in the notification a few macros, and all will be clear to the recipient.
I certainly understand your idea - that in an interface everything was great in one line (for example in events), but ... contact to developers

**aprudnev** · 09-11-2010, 02:34

Originally posted by zalex_ua

You do not accidentally forget about the macro {ITEM.NAME<1-9>} for notifications?

No, I did not. It can't be used in trigger name and then it is not exactly what we need (the name of the monitored object, not item). Some _similar_ macro can be a solution, true (moreover; it is the ONLY safe solution because $1 $2 etc are ALREADY used and can't be simple redefined) . For example. {ITEM.KEY<1-9>} can be some kind of solution (not so simple because we have item number and then parameter inside the item, and may have more then 1 item in the trigger, but $1 - $9 experience the same problem). Maybe {ITEM.KEY<1-9>[.ARG<1-9>]} . Ideally it should be allowed in trigger name(s).

**zalex_ua** · 09-11-2010, 02:59

Originally posted by aprudnev

Maybe {ITEM.KEY<1-9>[.ARG<1-9>]} . Ideally it should be allowed in trigger name(s).

very, very nice idea, really. But i think like {ITEM.KEY<1-9>.ARG<1-9>} is more optimal.
See my related issue request ZBX-2756

ok. now 3.00 am, so i'm go to sleep.

**aprudnev** · 09-11-2010, 03:20

Originally posted by zalex_ua

very, very nice idea, really. But i think like {ITEM.KEY<1-9>.ARG<1-9>} is more optimal.
See my related issue request ZBX-2756

ok. now 3.00 am, so i'm go to sleep.

yes, your request looks correct except 2 improvements - add .ARG<N> and ask if it can be used in trigger names (at least will be expanded in the notification).

Ad Widget

My (professional) feedback

My (professional) feedback

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment