I accidentally discovered a problem with the alarm sounds played by midas.

The javascript code is very simple: var audio=new Audio("alarm.mp3"); audio.play();

In the past, this reliably played a sound.

More recently, I started seeing javascript log messages about "unhandled exception" from audio.play(). (for me, they often
interrupt my javascript debugger sessions, in a very annoying way).

Adding an exception handler to audio.play() was not effective: it turns out, these days, audio.play() returns a Promise
and one must handle the Promise rejection case. (also it turns out that in the rejection handler
one *must* clear audio.src to avoid problems with excessive memory and cpu use).

But why does audio.play() throw a rejected promise?!?

This is the error message provided: "play() failed because the user didn't interact with the document first. https://goo.gl/xX8pDD"
name: "NotAllowedError".

The link takes us to https://developers.google.com/web/updates/2017/09/autoplay-policy-changes. This web page seems to explain
everything neatly, but most of the information turns out to be wrong and unhelpful:

a) "Muted autoplay is always allowed" - wrong - I see audio.play() rejection from muted tabs
b) "User has interacted with the domain (click, tap, etc.)." - wrong - I see rejection in the javascript debugger even as I debug the web page
c) "user's Media Engagement Index threshold has been crossed" - broken - playback count and MEI is always 0 (zero) for midas - per chrome://media-engagement/
d) "user has added the site to their home screen on mobile or installed the PWA on desktop" - what ?!?
e) chrome://flags/#autoplay-policy does not exist (was removed)

There may be a way to start google-chrome with special flags to globally enable autoplay, but it seems
to be aimed at "kiosk" type applications, not for general use browsers:
https://stackoverflow.com/questions/57455849/chrome-autoplay-policy-chrome-76

The bottom line:

there is no way to ensure that the alarm sound will always play

K.O.

P.S. I am tracking this problem here:
https://bitbucket.org/tmidas/midas/issues/191/exception-on-audioplay

> > Hi, Stefan - I did not look at your code, if all midas tabs are inactive, will the alarm sound still play?
> I added some code to do ONLY alarm updates if in the background (once every 10 seconds)

Ok, good.

> I don't have however a solution to your problem "not enough interaction with page" which I have never seen so far.

When it happens, you should see my messages about it in the javascript console. (rejected promise from audio.play()).

How to make it happen so one can see it and be sure it is handled correctly (audio.src has to be set to blank, at the least), I have no idea.

What I have been looking at for the huge memory and cpu use problem, I would call "a bad interaction between two kludges",
the first kludge is the throttling javascript in inactive tabs, the second kludge is the "not interacted" rejection of audio.play() promise.
(audio.play() returning a promise, and the strange sequencing between this promise success/reject and
the audio events loaded/canplay/done/etc smells of a kludge, too).

> I wonder if we should try push notifications: https://developers.google.com/web/fundamentals/codelabs/push-notifications
> which seems a bit complicated to me. It might also be too subtle if someone is sleeping in front of the computer.

I did not read all the explanation, but if it requires use of 3rd party services, I think we cannot use it, we do not want
to miss "experiment is on fire" alarms just because google is down.

K.O.

> a) google chrome slows down the execution of javascript in inactive tabs, leading to trouble
> with memory management - midas pages poll at 1/sec, each poll allocates memory for processing RPC messages,
> and (until recently) allocates memory for new DOM objects to update the web page - but the garbage collector
> gets slowed down and does not keep up - leading to huge memory use (up to 200 Mbytes) for inactive midas pages
> that normally consume 50-ish Mbytes.

Try my latest change, which drops any update if hidden, except the alarm sound. At the moment this only works on the main "status" page.

> most midas web pages poll in two places - mhttpd_refresh() updates the current date timestamp, alarms, currently active midas.log message;
> and each page has it's own loop for updating it's own data (i.e. "alarms" page, "programs" page).

That's correct. I changed mhttpd_refresh() now and the main loop for the "status" page. If that works for everybody, we can do that also for "programs" and other pages. The code is here:

      if (document.hidden) {
         // don't update page if hidden
         setTimeout(update_page, 500);
         return;
      }

where "update_page" has to be replaced with the proper function.

Stefan

> Hi, Stefan - I did not look at your code, if all midas tabs are inactive, will the alarm sound still play?

Nope. All updates are done in mhhtpd_refresh(), and I changed it such that nothing is updated if hidden.

I agree however that this is bad. You want to hear alarms always. So I added some code to do ONLY alarm updates if in the background. 
No GUI changes, only audio playing. Checking is reduced from 1 Hz to 0.1 Hz (once every 10 seconds). I don't have however a solution 
to your problem "not enough interaction with page" which I have never seen so far.

I wonder if we should try push notifications: https://developers.google.com/web/fundamentals/codelabs/push-notifications
which seems a bit complicated to me. It might also be too subtle if someone is sleeping in front of the computer.

Stefan

> Very good idea. And thanks for finding the document.hidden solution. I put it in, so give it a try.

Hi, Stefan - I did not look at your code, if all midas tabs are inactive, will the alarm sound still play?

K.O.


> 
> Best,
> Stefan
> 
> 
> > I am currently testing the new history system on the mhttpd side and stumbled over the following issue: typically our user open a lot of midas web-page tabs and keep them open. With the current version this leads after a night typically to a state where the browser is busy with itself and not reacting anymore.
> > 
> > One important reason seems to be that ALL tabs trying to communicate all the time which is totally unnecessary, since I think a hidden tab should stay in a sleeping mode. 
> > 
> > I was browsing if there is a way to find out if a tab is active or not, and found the following API which exactly does this:
> > 
> > https://developer.mozilla.org/en-US/docs/Web/API/Page_Visibility_API
> > 
> > Furthermore, the simple
> > 
> > document.hidden 
> > 
> > tag, could be used to find out if the page is currently active.
> > 
> > Wouldn't it a good idea to send all midas tabs which are not active into a sleep mode and only reactivate them if they come into focus?
> > 
> > I had a quick look at the JavaScript libs of midas, but I am not quite certain where to best inject this. 

> I am currently testing the new history system on the mhttpd side and stumbled over the following issue:
> typically our user open a lot of midas web-page tabs and keep them open. With the current version this leads after a night typically to a state where the browser is busy with itself and not reacting anymore.
> 
> One important reason seems to be that ALL tabs trying to communicate all the time which is totally unnecessary, since I think a hidden tab should stay in a sleeping mode. 

I am looking at two more problems with inactive tabs:

a) google chrome slows down the execution of javascript in inactive tabs, leading to trouble
with memory management - midas pages poll at 1/sec, each poll allocates memory for processing RPC messages,
and (until recently) allocates memory for new DOM objects to update the web page - but the garbage collector
gets slowed down and does not keep up - leading to huge memory use (up to 200 Mbytes) for inactive midas pages
that normally consume 50-ish Mbytes.

b) the playing of the alarm sound is throttled by "user has not interacted with the document" thing, but there is a bug -
instead of canceling the playing of the alarm sound, the sound file is still loaded, (but not played). (this is hard to debug
because I do not know how to manually trigger the "user has not interacted..." condition, I have to wait for many days for it.
Then, for inactive tabs, the loading of the sound files is slowed down, leading to many of them getting queued up,
and eventually they all try to load and play at the same time, again leading to huge memory and cpu use in inactive tabs.
(this sounds incredible because we play the alarm sound at most 1/minute, for sure the previous sound file must have
finished playing by then, but no, it is easy to see it happen - add a few console.log messages and wait for a few days).

>
> I was browsing if there is a way to find out if a tab is active or not, and found the following API which exactly does this:
> https://developer.mozilla.org/en-US/docs/Web/API/Page_Visibility_API
> 

From looking at the inactive tab business, I see that javascript in inactive tabs runs quite differently from javascript
in active tabs (i.e. timers do not work the same) and I see how the "visibility api" had to be invented to counter that.

> 
> Wouldn't it a good idea to send all midas tabs which are not active into a sleep mode and only reactivate them if they come into focus?
> I had a quick look at the JavaScript libs of midas, but I am not quite certain where to best inject this.
>

most midas web pages poll in two places - mhttpd_refresh() updates the current date timestamp, alarms, currently active midas.log message;
and each page has it's own loop for updating it's own data (i.e. "alarms" page, "programs" page).

we should be careful to not completely disable all polling as some experiments do use and do rely on the midas producing
loud alarm messages ("program logged is not running!!!", "program mhttpd aborted!!!"). Even if all midas tabs are inactive,
some javascript is some tab still has to run frequently enough to poll midas and to sound the alarm sounds (even though
I am not sure how to 100% reliably counteract the google-chrome not playing sound files because
of the "user did not interact with the site..." thing).

K.O.

> It is reproducible alright.

Thanks. At first blush, a guess, read_passwords() is not thread-safe and is called from multiple threads, not protected by semaphore. Crash report shows 2 active threads 
(one made it is far as processing the mjson rpc, the other one crashed in read_passwords()).

K.O.


> Here are the core dump and the backtrace (I think  the former is more informative).
> 
> 
> 
> > > Context: out of the box  MIDAS (using cmake) on OSX Mojave. 
> > > 
> > > Running with mongoose/opensslm installation following instruction here:
> > > https://midas.triumf.ca/MidasWiki/index.php/Quickstart_Linux
> > > 
> > > mhttpd crashing when midas webpage opened with Safari (12.1.2). Usually when opening the "chat" tab but sometimes also with the "message" tab.
> > > mhttpd(11109,0x70000827a000) malloc: *** error for object 0x7f8669501ef0: pointer being freed was not allocated
> > > mhttpd(11109,0x70000827a000) malloc: *** set a breakpoint in malloc_error_break to debug
> > > 
> > > No crash if using firefox (70.0.1 (64-bit))
> > 
> > I think we also have reports of mhttpd crash on macos with safari from the Dragon experiment,
> > but cannot reproduce the problem.
> > 
> > If you can reproduce this, can you capture the crash stack trace?
> > 
> > One way to do this is to enable core dumps in odb "/expt/enable core dumps" set to "y", restart mhttpd,
> > wait for the crash. I think macos writes core dumps into /cores/... Or you can run mhttpd inside lldb
> > and wait for the crash. the lldb command to show the stack trace is "bt", but you may need
> > to switch to different threads to see which one actually crashed. I forget what the command
> > for that is.
> > 
> > BTW, the mhttpd networking code has not changed in a long time, but an update
> > of mongoose web server library is overdue (to fix a memory leak, at least).
> > 
> > K.O.

Very good idea. And thanks for finding the document.hidden solution. I put it in, so give it a try.

Best,
Stefan


> I am currently testing the new history system on the mhttpd side and stumbled over the following issue: typically our user open a lot of midas web-page tabs and keep them open. With the current version this leads after a night typically to a state where the browser is busy with itself and not reacting anymore.
> 
> One important reason seems to be that ALL tabs trying to communicate all the time which is totally unnecessary, since I think a hidden tab should stay in a sleeping mode. 
> 
> I was browsing if there is a way to find out if a tab is active or not, and found the following API which exactly does this:
> 
> https://developer.mozilla.org/en-US/docs/Web/API/Page_Visibility_API
> 
> Furthermore, the simple
> 
> document.hidden 
> 
> tag, could be used to find out if the page is currently active.
> 
> Wouldn't it a good idea to send all midas tabs which are not active into a sleep mode and only reactivate them if they come into focus?
> 
> I had a quick look at the JavaScript libs of midas, but I am not quite certain where to best inject this. 

It is reproducible alright.
Here are the core dump and the backtrace (I think  the former is more informative).



> > Context: out of the box  MIDAS (using cmake) on OSX Mojave. 
> > 
> > Running with mongoose/opensslm installation following instruction here:
> > https://midas.triumf.ca/MidasWiki/index.php/Quickstart_Linux
> > 
> > mhttpd crashing when midas webpage opened with Safari (12.1.2). Usually when opening the "chat" tab but sometimes also with the "message" tab.
> > mhttpd(11109,0x70000827a000) malloc: *** error for object 0x7f8669501ef0: pointer being freed was not allocated
> > mhttpd(11109,0x70000827a000) malloc: *** set a breakpoint in malloc_error_break to debug
> > 
> > No crash if using firefox (70.0.1 (64-bit))
> 
> I think we also have reports of mhttpd crash on macos with safari from the Dragon experiment,
> but cannot reproduce the problem.
> 
> If you can reproduce this, can you capture the crash stack trace?
> 
> One way to do this is to enable core dumps in odb "/expt/enable core dumps" set to "y", restart mhttpd,
> wait for the crash. I think macos writes core dumps into /cores/... Or you can run mhttpd inside lldb
> and wait for the crash. the lldb command to show the stack trace is "bt", but you may need
> to switch to different threads to see which one actually crashed. I forget what the command
> for that is.
> 
> BTW, the mhttpd networking code has not changed in a long time, but an update
> of mongoose web server library is overdue (to fix a memory leak, at least).
> 
> K.O.

I am currently testing the new history system on the mhttpd side and stumbled over the following issue: typically our user open a lot of midas web-page tabs and keep them open. With the current version this leads after a night typically to a state where the browser is busy with itself and not reacting anymore.

One important reason seems to be that ALL tabs trying to communicate all the time which is totally unnecessary, since I think a hidden tab should stay in a sleeping mode. 

I was browsing if there is a way to find out if a tab is active or not, and found the following API which exactly does this:

https://developer.mozilla.org/en-US/docs/Web/API/Page_Visibility_API

Furthermore, the simple

document.hidden 

tag, could be used to find out if the page is currently active.

Wouldn't it a good idea to send all midas tabs which are not active into a sleep mode and only reactivate them if they come into focus?

I had a quick look at the JavaScript libs of midas, but I am not quite certain where to best inject this. 

> Context: out of the box  MIDAS (using cmake) on OSX Mojave. 
> 
> Running with mongoose/opensslm installation following instruction here:
> https://midas.triumf.ca/MidasWiki/index.php/Quickstart_Linux
> 
> mhttpd crashing when midas webpage opened with Safari (12.1.2). Usually when opening the "chat" tab but sometimes also with the "message" tab.
> mhttpd(11109,0x70000827a000) malloc: *** error for object 0x7f8669501ef0: pointer being freed was not allocated
> mhttpd(11109,0x70000827a000) malloc: *** set a breakpoint in malloc_error_break to debug
> 
> No crash if using firefox (70.0.1 (64-bit))

I think we also have reports of mhttpd crash on macos with safari from the Dragon experiment,
but cannot reproduce the problem.

If you can reproduce this, can you capture the crash stack trace?

One way to do this is to enable core dumps in odb "/expt/enable core dumps" set to "y", restart mhttpd,
wait for the crash. I think macos writes core dumps into /cores/... Or you can run mhttpd inside lldb
and wait for the crash. the lldb command to show the stack trace is "bt", but you may need
to switch to different threads to see which one actually crashed. I forget what the command
for that is.

BTW, the mhttpd networking code has not changed in a long time, but an update
of mongoose web server library is overdue (to fix a memory leak, at least).

K.O.

Context: out of the box  MIDAS (using cmake) on OSX Mojave. 

Running with mongoose/opensslm installation following instruction here:
https://midas.triumf.ca/MidasWiki/index.php/Quickstart_Linux

mhttpd crashing when midas webpage opened with Safari (12.1.2). Usually when opening the "chat" tab but sometimes also with the "message" tab.
mhttpd(11109,0x70000827a000) malloc: *** error for object 0x7f8669501ef0: pointer being freed was not allocated
mhttpd(11109,0x70000827a000) malloc: *** set a breakpoint in malloc_error_break to debug

No crash if using firefox (70.0.1 (64-bit))

> > > The compile option -DHAVE_FTPLIB checked in mdsupport.cxx disappeared if you 
> > > compile with cmake.
> > 
> > Hi, Stefan - do we still need to support FTP in the logger? In the lazylogger, special support for 
> > FTP is not needed, they can you the "script" method and do FTP without our help.
> > 
> > I move to remove FTP support from MIDAS. (second? other opinions?)
> 
> I oppose to remove FTP support from lazylogger.

Confirmed. FTP support in lazylogger stays.

K.O.

> I keep on getting messages like this:
> 16:25:35 [fecaen,ERROR] [odb.c:4567:db_get_data,ERROR] data for key
> "/DAQ/params/VX1730/custom/Board 0/Channel 0/Input range" truncated
>
>  [  bool fInputRange... ]
>          size = sizeof(fInputRange);
>          db_get_data(hDb, hSubKey, &fInputRange, &size, TID_BOOL);
>

The error is correct. size of TID_BOOL is 4 byte (uint32_t) and you give is sizeof(bool) instead which is probably not 4.

Note that sizeof(bool) is not well defined, sometimes it is 1 (you need 4), sometimes something else, see
https://stackoverflow.com/questions/4897844/is-sizeofbool-defined-in-the-c-language-standard

A good fix would be to change fInputRange from bool to uint32_t (which is always 4 byte size).

#include <stdint.h>
...
uint32_t fInputRange;

K.O.

I keep on getting messages like this:

16:25:35 [fecaen,ERROR] [odb.c:4567:db_get_data,ERROR] data for key
"/DAQ/params/VX1730/custom/Board 0/Channel 0/Input range" truncated

whenever I start my frontend. Input range is defined to be a BOOL and using
odbedit to read it shows:

Key name                        Type    #Val  Size  Last Opn Mode Value
---------------------------------------------------------------------------
Input range                     BOOL    1     4     75h  0   RWD  y

without any error message. The entry is read using

         size = sizeof(fInputRange);
         db_get_data(hDb, hSubKey, &fInputRange, &size, TID_BOOL);

where fInputRange is a bool.

Where does this message come from and how can I resolve this?

I have written a system monitor tool for MIDAS, that has been merged in the develop branch today: sysmon

https://bitbucket.org/tmidas/midas/pull-requests/8/system-monitoring-a-new-frontend-to-log/diff

To use it, simply run the new program
sysmon
on any host that you want to monitor, no configuring required.




The program is a frontend for MIDAS, there is no need for configuration, as upon initialisation it builds a history display for you. Simply run one instance per machine you want to monitor. By default, it only logs once per 10 seconds.

The equipment name is derived from the hostname, so multiple instances can be run across multiple machines without conflict. A new history display will be created for each host.

sysmon uses the /proc pseudo-filesystem, so unfortunately only linux is supported. It does however work with multiple architectures, so x86 and ARM processors are supported.

If the build machine has NVIDIA drivers installed, there is an additional version of sysmon that gets built: sysmon-nvidia. This will log the GPU temperature and usage, as well as CPU, memory and swap. A host should only run either sysmon or sysmon-nvidia

elog:1727/1 shows the History Display generated by sysmon-nvidia. sysmon would only generate the first two displays (sysmon/localhost and sysmon/localhost-CPU)

There is a not-so-well-known function in the ODB to write protect some keys. You can do

odbedit> chmod 1 /Equipment/HV/Demand

which will write protect your Demand values. You see that by doing

odbedit> ls -ls

where you only see a "R" at the right end instead a "RWD". I haven't tried it yet (so better do a dry run yourself), but that should prevent an odb load to overwrite your demand values. To change the values, put some logic on a custom page to unprotect the 
values, change them, and then protect them again.

Stefan

Hi,

I have been looking at the 2019 workshop slides, I am interested in the C++ future of MIDAS. 

I am quite interested in using the object oriented 


ALPHA will start data taking in 2021

> > The compile option -DHAVE_FTPLIB checked in mdsupport.cxx disappeared if you 
> > compile with cmake.
> 
> Hi, Stefan - do we still need to support FTP in the logger? In the lazylogger, special support for 
> FTP is not needed, they can you the "script" method and do FTP without our help.
> 
> I move to remove FTP support from MIDAS. (second? other opinions?)

I oppose to remove FTP support from lazylogger. We still use it heavily at PSI. In comparison to the "script" method, it 
shows the current speed in MB/s which helps us to diagnose some network problem by writing this number into the 
history. The "script" method only give you an integral transfer speed after a file has be completely written.

I'm however not sure who FTP is used in lazylogger. It goes into mdsupport.cxx and I seem to remember that Pierre 
wrote the FTP code by hand, so no external library is necessary.

Stefan

> Yes, we could have
> kept that apart, yes, in this case a double would also work (and not break things), but a bug is a bug...
> I could think of senisble use cases where doubles and ints are mixed and I also know quite a few areas where it makes
> sense to use floats...

I agree with Nik that we should fix this on the midas level. Since it happens in history_schema.cxx which was written by KO, maybe he can have a look.

Stefan