PRDCR_UNIQUE_DCU_REPORTED_PER_S -> the number of dcus that reported at least once in the last second
PRDCR_TOTAL_DCU_REPORTED_PER_S -> the count of dcus that reported in the last second.  In good conditions this should be 16*PRDCR_UNIQUE_DCU_REPORTED_PER_S

PRDCR_TOTAL_DATA_RATE_KB_PER_S -> the total data rate over the last second in kb (excludes < 1kb of headers a second)
PRDCR_TP_DATA_RATE_KB_PER_S -> the data rate of test point data over the last second in kb
PRDCR_MODEL_DATA_RATE_KB_PER_S -> the data rate of the regular model data over the last second in kb

PRDCR_OPEN_TP_COUNT -> the count of test points opened sampled once per second

added some unit tests for checkdaqconfig.py.  Now accepts hardcoded paths in master template, but will look in expected paths if the paths in the file are dummies (template variables for example).

Erik von Reis authored 4 years ago and Jonathan Hanks committed 4 years ago

adding script for archiving daq configs when using multiple DAQs.  Script uses a hash to dermine when to update archives.  Manages separte pointers for each DAQ.

required to compile IOP to run at 128K and user apps to receive
data at this higher rate.

control models running with the higher rate IOP.  To make this change,
certain defines were moved from header files to the feCodeGen itself
to avoid introducing more ifdefs in the code.

built with no I/O chassis flag.

Update the test to check if daqd stops on a timestamp jump to give a proper return value on success.

changed heartbeat loop to eliminate race condition where end-of-epoch test for epoch 15 might succeed immediately, triggering rapid heartbeats every 10 ms for the next 60+ ms.

The stopping_now_ flag should be set to true to signal trender threads to stop.

Also move some stack based variable sized arrays onto the heap in std::vectors to remove a place that the stack can be overflowed.

Added a test to ensure the daqd stops when there is a jump in the data, and updated the live test daqdrc to start the trender thread to highlight issues with stopping.

Historically, there have been two main types of broadcasts in daqd, the DC data stream (the 10Gb udp broadcast) and the GDS frame transfer.  When DATA_CONCENTRATOR was defined only the DC data stream is used.  This mode has been removed from the daq architecture.  So the only remaining broadcast mode should be the GDS frame transfer.

The big difference between the two broadcast methods is the data source.  For the DC stream the source is the main circular buffer.  For the GDS the data source is a frame file that the framer_io function mmaps and pushes as a block of memory.

There were several modes of operation for the daqd, and DATA_CONCENTRATOR was picked when collapsing the daqd down to one binary.  So remove support for the DC data stream.
 * Removes support for broadcasting testpoints
 * Removes the udp broadcast logic for the DC stream
 * Removes a number of #if DATA_CONCENTOR macro guards
 * Removes a number of #if GDS_TESTPOINT macro guards

 The final flow of the broadcast data is now as follows:

 the parser sees the request for the net-writer with a broadcast address.
 This triggers a net writer to be created, a producer thread to start, a circular buffer to be created.  The producer copies data from the main circular buffer and unlocks the main circular buffer it puts into the new circular buffer with a nowait put operation (which is allowed to fail).  Nothing has ever emptied this, so a consumer thread is not launched (it didn't do anything anyways).

From the user perspective, key changes are in the model Parameter block settings:
- An IOP model should now be designaeted with iop_model=1
	- Previously set by adcMaster=1
- A user control application should have the adcSlave=1 parameter removed.
	- If iop_model=1 is not in the Parameter block, then RCG assumes
	  this is a user control application model.
- IOP models that send/rcv timing over the Dolphin network
	- The transmitter model Parameter block needs dolphin_time_xmit=1
	- A recevier IOP model Parameter block needs dolphin_time_rcvr=1

Work to remove slave from the code.  This is not a complete cleanup, but it goes through

 * src/daqd
 * src/gds
 * src/gdsConfig
 * src/mx_stream

Change an || to an &&, so the loop that waits for work or a stop signal exits when one of those conditions is met.

hangups for up to 61msec.

Erik von Reis authored 4 years ago and Jameson Rollins committed 4 years ago

awgtpman now prefers to use the TARGET environment variable for the target directory.
If not present, /opt/rtcds/<site>/<ifo> will be used.
If neither TARGET nor SITE and IFO are set, awgtpman will refuse to load
and print a helpful error.

awgtpman will look in $TARGET/target/gds/param for the parameter files
to get the model rate.
awgtpman will now fail with a helpful error when the model rate cannot be found.
If the '-r' option is set, then awgtpman will uses exclusively the command line arguments to
determine model rate and won't check the param files.
awgtpman will ignore command line arguments for setting the model rate unless '-r' is given.

 * Spelling fix, changing to stop_function_t from stop_funtion_t.
 * Corrected the timed circular buffer get to timeout.

Move all the trender threads to being tracked by a thread handler and having a uniform stop signal.

Add a timed_get function to the circular buffer which will allow queries to the circualr buffers to time out.  This is used to allow the trender to see a stop request.

Converting it to use atomics.

This is done as part of expanding the wrapping of the pthread... calls with C++.  Starting on a interface to better track threads.

Added back in site_prefix and archive.  Removed awgapi.h and awgapi.c for being unused.  cleaned up readChnInfo() in gdschannel.c, but it now no longer tries to open a non-existant channel database file.

Also removes some other unused variables.

The buffer enters into a late state after its data has been read out.  Any data for that cycle that arrives is considered late, until the next cycle comes in, at which point it is marked as dropped.

This updates the late state when data is copied out of the buffer and clears it when a new gps second/cycle is accepted in.

Closes #119.

It will show drops in dcus (counts).

Erik von Reis authored 4 years ago and Jameson Rollins committed 4 years ago

closes #116

Main issue was extra space in my ifdef TIME_SLAVE in time_slave_io.c,
therefore code not compiled correctly and reverted to internal timer.