Difference between revisions of "Stateframe Database"

From EOVSA Wiki
Jump to: navigation, search
(Initial setup of database access software)
Line 1: Line 1:
  
 
== Setup and Initial Test of Python Access to Existing Database ==
 
== Setup and Initial Test of Python Access to Existing Database ==
 
  
 
=== Introduction ===
 
=== Introduction ===
Line 219: Line 218:
 
* <code>cursor = get_cursor()</code>  Opens the database and returns a cursor for access to it.
 
* <code>cursor = get_cursor()</code>  Opens the database and returns a cursor for access to it.
 
* <code>mydict = get_dbrecs(cursor, version, dimension, timestamp, nrecs)</code>  Takes as input an open cursor and version, dimension, timestamp and nrecs, and returns a dictionary from the table indicated by the version and dimension, starting at timestamp, and having nrecs entries.  The data in each mydict key has dimensions of dimension x nrecs.
 
* <code>mydict = get_dbrecs(cursor, version, dimension, timestamp, nrecs)</code>  Takes as input an open cursor and version, dimension, timestamp and nrecs, and returns a dictionary from the table indicated by the version and dimension, starting at timestamp, and having nrecs entries.  The data in each mydict key has dimensions of dimension x nrecs.
 +
 +
== Useful Tricks ==
 +
=== Get Every Nth Entry ===
 +
To get every Nth entry, for example, one data point each hour, use the modulo (%) function.  For example, this gets Ant 14's VPol power once an hour:
 +
<pre>query = 'select Timestamp,Ante_Fron_FEM_VPol_Power from fV66_vD15 where (I15 % 15) = 13 and
 +
Timestamp between '+t0+' and '+t1+' and (cast(Timestamp as bigint) % 3600) = 0 order by Timestamp'
 +
</pre>
 +
where t0 is the start Labview timestamp as a string, and t1 is the end Labview timestamp as a string.

Revision as of 17:24, 25 April 2017

Setup and Initial Test of Python Access to Existing Database

Introduction

This document describes the initial setup and access of the EOVSA Stateframe database from Python. The original version of this document described the case of a database called “eOVSA05” that existed on a machine named “GARY-FY13NB” (laptop). That connection string for connecting from the laptop was:

cnxn = pyodbc.connect(“DRIVER={SQL Server}; SERVER=GARY-FY13NB; DATABASE=eOVSA05;Trusted_connection=yes;")

However, some adjustments were needed to allow connection to the actual server machine (EOVSASQL), using an IP address and “SQL Authentication.” The trick was to create a user on the server in the SQL Server Management Studio (user name is ‘aaa’, password is ‘I@bsbn2w’), and then use the server IP address (currently ‘128.235.89.168’) on port 1433. This port had to be opened in the server’s Windows Firewall. The new connection string is:

cnxn = pyodbc.connect("DRIVER={SQL Server}; SERVER=128.235.89.168,1433; DATABASE=eOVSA06;UID=aaa;PWD=I@bsbn2w;")

The other commands have been updated for the current situation, also. As of 2014-Sep-30, the SQL Server has been set up at OVRO and the connection string from Helios is:

cnxn = pyodbc.connect("DRIVER={FreeTDS}; SERVER=192.168.24.106,1433; DATABASE=eOVSA06;UID=aaa;PWD=I@bsbn2w;")


Initial setup of database access software

The Python library to access Microsoft SQL Server is called pyodbc. This describes downloading the library and integrating it with an existing python2.76 installation, making a first connection to the database, and executing a query. The query assumes that database “eOVSA06” has a populated table “fV26_vD15.”

1. First install pyodbc ([1]). For instance, pyodbc-3.0.7.win-amd64-py2.7.exe.

2. Start ipython and type

import pyodbc

3. Connect to database with

cnxn = pyodbc.connect("DRIVER={SQL Server}; SERVER=128.235.89.168,1433; DATABASE=eOVSA06; UID=aaa; PWD=I@bsbn2w;")

4. Get a 'cursor' to the database:

cursor = cnxn.cursor()

5. Send an SQL query:

cursor.execute("select top 16 * from fV26_vD15 order by Timestamp")

6. Fetch the data returned by the query

rows = cursor.fetchall()

7. rows now contains a list of 16 rows.


Note that setup on Linux (Ubuntu) is quite a bit more complicated. Here are the steps:

1. Download pyodbc-3.0.7.zip from the above link (or copy from helios.solar.pvt).

2. Install unixODBC and FreeTDS packages:

sudo apt-get install unixODBC-dev freetds-dev freetds-bin tdsodbc

3. Edit /etc/freetds/freetds.conf (or copy from helios) to change the last 4 lines to

        [sqlserver]
	host = sqlserver.solar.pvt
	port = 1433
	tds version = 11.0

4. Edit file /etc/odbc.ini (or copy from helios) to contain:

        [sql]
        Driver = FreeTDS
	Description = ODBC connection via FreeTDS
	Trace = No
	Servername = sqlserver.solar.pvt
	Database = eOVSA06

5. Edit file /etc/odbcinst.ini (or copy from helios) to contain:

        [FreeTDS]
	Description     = TDS driver (Sybase/MS SQL)
	Driver = /usr/lib/x86_64-linux-gnu/odbc/libtdsodbc.so
	Setup = /usr/lib/x86_64-linux-gnu/odbc/libtdsS.so
	CPTimeout       =
	CPReuse         =
	FileUsage       = 1

6. Unzip the zip file in step 1, cd to the directory pyodbc-3.0.7, and type:

	
        python setup.py build
	sudo python setup.py install

From there, use the connection string for Linux shown in the introduction section, and proceed from there.

Examples of returned queries

cursor.execute(“select top 2 * from fV26_vD1 order by Timestamp”)
rows = cursor.fetchall()

Returns the entire “dimension-1” data for the first two entries in the table, ordered by Timestamp. The contents can be accessed one row at a time (rows[0] and rows[1] in this case), and the entry names can be listed with rows[0].cursor_description. One can access the value of, say, the current temperature via rows[0].Sche_Data_Weat_Temperature:

cursor.execute(“””select top 20 * from fV26_vD15 a where (a.[I15] % 15) = 0 order by Timestamp”””)
rows = cursor.fetchall()

Returns the first 20 rows for Antenna 1 [via (a.[Index] % 15) = 0] of the dimension-15 table, ordered by Timestamp. One can check that these are all for the same antenna, for example, by checking one of the pointing coefficients using the Python code:

for row in rows: print row.Ante_Cont_PointingCoefficient2


Create a new StateFrameDef Table entry

Whenever a new version of the stateframe is created due to some change in the content of the stateframe, a new StateFrameDef table entry must be created. This is accomplished via a set of SQL commands, like:

cursor.execute(“insert into StateFrameDef (Status, Version, Dimension, DataType,  FieldBytes,  DimOffset,  StartByte,  FieldNum,  FieldName) 
                    values ( 0, 15, 15, 'u16', 2, 28, 759, 1, 'DCM_Slot')”)

and so on for the entire new table. Once all of the rows of the table are entered, the command to create the tables for that version is:

cursor.execute(“update StateFrameDef set status=1 where Version=’15’”) 

NB: Once a stateframedef entry is entered, it cannot be entered again within causing unrecoverable errors in the table. See below for description of procedure to clear all of the tables and start ove


Complete process of inserting a new binary record

The structure of binary data for a given stateframe version is encoded in its XML file with the name ‘stateframe_vxx.00.xml’, where xx is the version number, i.e. 26. The binary data has to be rearranged via a routine in stateframedef.py called transmogrify(). A complete recipe for reading and inserting data from a stateframe log file (‘sf_20140205_v26.0.log’ in this example), is as follows:

import stateframedef as sfd
import pyodbc
sf, version = sfd.rxml.xml_ptrs(‘stateframe_v26.00.xml’)
brange, outlist = sfd.sfdef(sf)
f = open(‘sf_20140205_v26.0.log’,’rb’)
buf = f.read(32)
recsize = sfd.struct.unpack_from(‘i’, buf, 16)[0]
f.close()
cnxn = pyodbc.connect("DRIVER={SQL Server};SERVER=128.235.89.168,1433; 
DATABASE=eOVSA06;UID=aaa;PWD=I@bsbn2w;")
cursor = cnxn.cursor()
with open(‘sf_20140205_v26.0.log’, ‘rb’) as f:
    bufin = f.read(recsize)
    bufout = sfd.transmogrify(bufin, brange)
    try:
        cursor.execute(‘insert into fBin (Bin) values (?)’,  
        pyodbc.Binary(bufout))
    except:
              pass
cnxn.commit()

The execute line may return an error and fail, of course, especially if the data have previously been inserted, so it is enclosed in a try: except: clause so that everything does not immediately stop. Error checking would go into the except: clause.


Procedure to create the information for a new stateframedef table

The Python code in stateframedef.py does all of the manipulation related to creating StateFrameDef (and also ScanHeaderDef) tables as well as converting stateframe (or scanheader) binary data to the database data. For various reasons involving the internals of SQL, it is necessary to reorder certain data (two-dimensional arrays) in the stateframe and scanheader before they can be saved in the SQL database. These special cases are handled in the bowels of stateframedef.py’s walk_keys() routine. To create a stateframedef table from a new stateframe xml file, e.g. stateframe_v32.00.xml, do:

import stateframedef as sfd
sf, version = sfd.rxml.xml_ptrs(‘stateframe_v32.00.xml’)
brange, outlist = sfd.sfdef(sf)
sfd.startbyte(outlist)
tbl = sfd.outlist2table(outlist,version)

This last line prints the table to the screen and also creates the contents of the table as commands for input to SQL. To upload the table in SQL and activate it, the commands are:

for line in tbl:
cursor.execute(line)
cursor.execute(“update StateFrameDef set status=1 where Version=’” +  
                str(int(version)) + ”’”)


To close a connection

Once a connection cxnx has been created, and a cursor has been defined, do the following to release them:

cursor.close()
del cursor
cnxn.close()


To clear all tables and start from scratch

The SQL tables that describe each version of the stateframe or scanheader are called (case-insensitive) StateFrameDef and ScanHeaderDef. If somehow these get confused (as in entering a line that is already in the table), any previously backed-up tables can be restored using:

cursor.execute(“ov_fTEST_DefRestore”) 

However, if no appropriate backup exists, the tables need to be cleared and reloaded. This is a very quick process, luckily. To empty the “definition” tables, execute the following:

cursor.execute(“ov_fTEST_DefTruncate”)

To reload them, it is intended to have a single function (not available yet):

reload_deftables(),

which will clear the tables and reload them automatically. Right now, there is a routine that does this for one file:

flag = load_deftable(xml_file),

which returns True if successful, or False if an error. If a definition for the specified version already exists in the table, a warning is generated and the table is not redefined, but the routine returns True. This avoids trying to redefine the table and thus messing it up. The reload_deftables() routine will just clear the tables, and then take all xml files in a directory and repeatedly call load_deftable().


Getting data across version boundaries

The data for each version of the stateframe appears in unique tables, so that the information for one period of time may be in, for example, fV32_vD15, while for the next adjacent time it is in fV35_vD15 (no data were recorded for versions 33-34). If one wants to get data that spans these two tables, one would use the following query (this example is for Ante_Cont_Elevation1):

cursor.execute(“””select Timestamp, Ante_Cont_Elevation1 
                  from fV32_vD15 a where (a.[i15] % 15) = 0 
                  union all 
                  select Timestamp, Ante_Cont_Elevation1 
                  from fV35_vD15 b where (b.[i15] % 15) = 0 
                                       order by TimeStamp”””)

Note that it is necessary to include in the select list the column (TimeStamp in this case) that is to be used for ordering the data. Otherwise one gets a cryptic error message [42000] ORDER BY items must appear in the select list if the statement contains a UNION, INTERSECT or EXCEPT operator. Once the data are selected, the following lines will plot them:

rows = cursor.fetchall()
elev = zeros(len(rows),'float')
times = zeros(len(rows),'float')
for i,x in enumerate(rows):
    times[i], elev[i] = x
plot(times-times[0],elev/10000.,'.')


New python code for accessing the database

There is a module called dbutil that contains (and will further be developed) routines to access the database. The two current routines are:

  • cursor = get_cursor() Opens the database and returns a cursor for access to it.
  • mydict = get_dbrecs(cursor, version, dimension, timestamp, nrecs) Takes as input an open cursor and version, dimension, timestamp and nrecs, and returns a dictionary from the table indicated by the version and dimension, starting at timestamp, and having nrecs entries. The data in each mydict key has dimensions of dimension x nrecs.

Useful Tricks

Get Every Nth Entry

To get every Nth entry, for example, one data point each hour, use the modulo (%) function. For example, this gets Ant 14's VPol power once an hour:

query = 'select Timestamp,Ante_Fron_FEM_VPol_Power from fV66_vD15 where (I15 % 15) = 13 and 
Timestamp between '+t0+' and '+t1+' and (cast(Timestamp as bigint) % 3600) = 0 order by Timestamp'

where t0 is the start Labview timestamp as a string, and t1 is the end Labview timestamp as a string.