ConditionsDbSchema: Difference between revisions
Line 7: | Line 7: | ||
==Calibration Tables== | ==Calibration Tables== | ||
Starting with tst.calib1, we have a conceptual content | Starting with tst.calib1, we have a conceptual content of a detector with 3 channels and calibraiotn which contains a integer flag and a float, "dtoe", calibration for each channel. | ||
{| class="wikitable" | {| class="wikitable" |
Revision as of 22:38, 24 October 2018
Introduction
A database schema is the design of the tables and functions, etc, and their interactions. Here we describe the conditions database in some detail.
We are using postgres databases. In postgres, the word "schema" is also used to refer to a what is essentially a folder, or a way to group database tables and assign permssions as a group. We have several of this type of schema, one for each detector subsystem (trk,cal,crv) plus one for the interval of validity structure (val) and one for test tables (tst). A table can be referred to as schema.name, for example, we have an example table in the tst schema, tst.calib1.
In this page we will use the word 'calibration to refer to one logical group of data, entered into one table at the same time. For example, if we had a set of gains for all crystals and committed those, with one value (equals one row) for each channel, that would comprise a "calibration". If the next week I committed another set of gains, that would be another "calibration".
Calibration Tables
Starting with tst.calib1, we have a conceptual content of a detector with 3 channels and calibraiotn which contains a integer flag and a float, "dtoe", calibration for each channel.
channel | flag | DtoE |
---|---|---|
0 | 12 | 1.11 |
1 | 13 | 2.11 |
2 | 11 | 3.11 |
This set of 3 rows are a logical group and were committed together, and so it is an example of a calibration.
In the database definition, this table needs to hold many calibrations and we need a way to separate them, so we add a column called cid, which stands for calibration ID. One cid number is unique across the whole database, so it labels exactly three rows in exactly this table. The actual database table looks like the following, with two calibrations entered, with cids of 1 and 2.
cid | channel | flag | DtoE |
---|---|---|---|
1 | 0 | 12 | 1.11 |
1 | 1 | 13 | 2.11 |
1 | 2 | 11 | 3.11 |
2 | 0 | 22 | 1.21 |
2 | 1 | 23 | 2.21 |
2 | 2 | 21 | 3.21 |
Intervals of Validity
Logically, an interval of validity (IOV) says that a specific cid (which implies one calibration in a specific table) is appropriate for an interval of subruns. There is another conceptual level which is not address yet. This level connects this IOV to a particular set of calibrations. For example, this IOV might belong in the first pass of production, but not in the second pass, which has a whole different set of calibrations. This logical connection is made below. In the IOV table here we only attach a subrun range to calibration. This can be re-used for many calibration sets.
IID | CID | Start run | Start subrun | End run | End subrun | create time | create user |
---|---|---|---|---|---|---|---|
1 | 1 | 1001 | 1 | 1001 | 999999 | 2018-10-12 08:58:26 | rlc |
2 | 2 | 1002 | 1 | 1004 | 1 | 2018-10-12 08:58:26 | rlc |
3 | 3 | 1004 | 2 | 999999 | 999999 | 2018-10-12 08:58:26 | rlc |