CyberShake Study Database Archiving

SQLite Archiving

The following procedure should be used when we want to archive a study to SQLite which is either in the production or data access database to disk, to free up room in the database.

First, dump the old database contents into a new directory by running the following commands.

Note that if you don't want to lock the tables, you can replace '--lock-all-tables' with '--single-transaction=TRUE' in the following commands.

1. Determine the Study ID you want to archive, from looking at the Studies table in the DB:
select * from Studies;
2. Dump the PeakAmplitudes, using:
mysqldump --lock-all-tables -u cybershk -p --where 'Run_ID in (select R.Run_ID from CyberShake_Runs R where R.Study_ID=<study id to archive>)' CyberShake PeakAmplitudes > peak_amps.sql
3. Dump Hazard_Datasets, using:
mysqldump --lock-all-tables -u cybershk -p --where 'Study_ID=<study id to archive>' CyberShake Hazard_Datasets > hazard_datasets.sql
4. Dump Hazard_Curves, using:
mysqldump --lock-all-tables -u cybershk -p --where 'Hazard_Dataset_ID in (select D.Hazard_Dataset_ID from Hazard_Datasets D where D.Study_ID=<study id to archive>)' CyberShake Hazard_Curves > hazard_curves.sql
5. Dump Hazard_Curve_Points, using:
mysqldump --lock-all-tables -u cybershk -p --where 'Hazard_Curve_ID in (select C.Hazard_Curve_ID from Hazard_Curves C, Hazard_Datasets D where D.Study_ID=<study id to archive>)' CyberShake Hazard_Curve_Points > hazard_curve_points.sql
6. Dump CyberShake_Runs, using:
mysqldump --lock-all-tables -u cybershk -p --where 'Study_ID=<study id to archive>' CyberShake CyberShake_Runs > runs.sql
7. You now need the rest of the input tables. They used to be pretty small, but now they're up to ~70GB with indices. As of the completion of Study 22.12, here are their approximate sizes (in the DB, not as dump files):

   Table	Size (GB)
   AR_Hazard_Curve_Points	44.6
   Points	6.3
   CyberShake_Site_Ruptures	10.3
   Ruptures	2.3
   Rupture_Variations	2.6
   AR_Hazard_Curves	1.4
   Other tables	0.5

   You can either:

   • Dump all of the input tables -- they are small enough that we don't mind capturing data which wasn't directly used in this study.
   mysqldump -u cybershk -p CyberShake AR_Hazard_Curve_Points AR_Hazard_Curves AR_Hazard_Datasets Atten_Rel_Metadata Atten_Rels CyberShake_Site_Regions CyberShake_Site_Ruptures CyberShake_Site_Types CyberShake_Sites ERF_IDs ERF_Metadata ERF_Probability_Models IM_Types Points Rupture_Variation_Probability_Modifier Rupture_Variation_Scenario_IDs Rupture_Variation_Scenario_Metadata Rupture_Variations Ruptures SGT_Variation_IDs SGT_Variation_Metadata Studies Time_Spans Velocity_Models > input_tables.sql
   • Dump only the sections of the large input tables that you need. Note that the Points table isn't used anymore, so don't bother to dump it.
     1. mysqldump --lock-all-tables -u cybershk -p --where 'Hazard_Curve_ID in (select C.Hazard_Curve_ID from Hazard_Curves C, Hazard_Datasets D where D.Study_ID=<study id to archive>)' CyberShake AR_Hazard_Curve_Points
     2. mysqldump --lock-all-tables -u cybershk -p --where 'CS_Site_ID in (select R.Site_ID from CyberShake_Runs R where R.Study_ID=<study id to archive>) and ERF_ID in (select R.ERF_ID from CyberShake_Runs R where R.Study_ID=<study id to archive>)' CyberShake CyberShake_Site_Ruptures > cs_site_ruptures.sql
     3. mysqldump --lock-all-tables -u cybershk -p --where 'ERF_ID in (select R.ERF_ID from CyberShake_Runs R where R.Study_ID=<study id to archive>) and (Source_ID, Rupture_ID) in (select distinct SR.Source_ID, SR.Rupture_ID from CyberShake_Site_Ruptures SR, CyberShake_Runs R where SR.CS_Site_ID=R.Site_ID and SR.ERF_ID=R.ERF_ID and R.Study_ID=<study id to archive>)' CyberShake Ruptures > ruptures.sql
     4. mysqldump --lock-all-tables -u cybershk -p --where '(Rup_Var_Scenario_ID, ERF_ID) in (select R.Rup_Var_Scenario_ID, R.ERF_ID from CyberShake_Runs R where R.Study_ID=<study id to archive>) and (Source_ID, Rupture_ID) in (select distinct SR.Source_ID, SR.Rupture_ID from CyberShake_Site_Ruptures SR, CyberShake_Runs R where SR.CS_Site_ID=R.Site_ID and SR.ERF_ID=R.ERF_ID and R.Study_ID=<study id to archive>)' CyberShake Rupture_Variations > rupture_variations.sql
     5. mysqldump --lock-all-tables -u cybershk -p --where 'AR_Hazard_Dataset_ID in (select distinct D.AR_Hazard_Dataset_ID from AR_Hazard_Datasets D, CyberShake_Runs R where R.Study_ID=12 and D.Time_Span_ID=1 and D.Prob_Model_ID=1 and D.ERF_ID=R.ERF_ID and D.Velocity_Model_ID=R.Velocity_Model_ID) and Lat>=(select distinct D.Min_Lat from AR_Hazard_Datasets D, CyberShake_Runs R where R.Study_ID=12 and D.Time_Span_ID=1 and D.Prob_Model_ID=1 and D.ERF_ID=R.ERF_ID and D.Velocity_Model_ID=R.Velocity_Model_ID) and Lat<=(select distinct D.Max_Lat from AR_Hazard_Datasets D, CyberShake_Runs R where R.Study_ID=12 and D.Time_Span_ID=1 and D.Prob_Model_ID=1 and D.ERF_ID=R.ERF_ID and D.Velocity_Model_ID=R.Velocity_Model_ID) and Lon>=(select distinct D.Min_Lon from AR_Hazard_Datasets D, CyberShake_Runs R where R.Study_ID=12 and D.Time_Span_ID=1 and D.Prob_Model_ID=1 and D.ERF_ID=R.ERF_ID and D.Velocity_Model_ID=R.Velocity_Model_ID) and Lon<=(select distinct D.Max_Lon from AR_Hazard_Datasets D, CyberShake_Runs R where R.Study_ID=12 and D.Time_Span_ID=1 and D.Prob_Model_ID=1 and D.ERF_ID=R.ERF_ID and D.Velocity_Model_ID=R.Velocity_Model_ID)' CyberShake AR_Hazard_Curves > ar_hazard_curves.sql
     6. mysqldump --lock-all-tables -u cybershk -p CyberShake AR_Hazard_Datasets Atten_Rel_Metadata Atten_Rels CyberShake_Site_Regions CyberShake_Site_Types CyberShake_Sites ERF_IDs ERF_Metadata ERF_Probability_Models IM_Types Rupture_Variation_Probability_Modifier Rupture_Variation_Scenario_IDs Rupture_Variation_Scenario_Metadata SGT_Variation_IDs SGT_Variation_Metadata Studies Time_Spans Velocity_Models

Next, convert the SQL dumps into SQLite format using mysql2sqlite:

1. For each of the 6 dump files, run:

 mysql2sqlite <SQL dump file> > <SQLite dump file>
 Example:  ./mysql2sqlite peak_amps.sql > peak_amps.sqlite

For large tables this may take an hour or two.

Create a SQLite database and import the tables. Use sqlite3 3.7.11 or later, or there will be an error reading the dump files.

1. Create a database for this study:
sqlite3 <study name>
2. For each dump file, run the following command:
.read <path/to/dump/file.sqlite>

Run a few queries on the original table on the database, and on the sqlite files to check that the count is the same.

1. Check the number of PeakAmplitudes:
select count(*) from PeakAmplitudes P, CyberShake_Runs R where R.Study_ID=<study_id> and P.Run_ID=R.Run_ID;
2. Check the number of rupture variations for each site:
select count(*) from CyberShake_Runs R, CyberShake_Site_Ruptures SR, Rupture_Variations V where R.Site_ID=SR.CS_Site_ID and SR.Source_ID=V.Source_ID and SR.Rupture_ID=V.Rupture_ID and R.ERF_ID=V.ERF_ID and R.ERF_ID=SR.ERF_ID and V.Rup_Var_Scenario_ID=R.Rup_Var_Scenario_ID and R.Study_ID=<study id>;

Create index on Run_ID and IM_Type_ID in PeakAmplitudes table to speed up access.

1. Copy over the original version to a new file.
2. Set the SQLITE_TMPDIR variable to point to a filesystem with lots of free space.
3. Using a current version of sqlite3, open the new database file and run: CREATE INDEX "idx_PeakAmplitudes_Run_ID_IM_Type_ID" ON "PeakAmplitudes" (`Run_ID`, `IM_Type_ID`); This may take several hours to finish.

Move the files to the archive location on project.

1. Tar up the sqlite files:
tar czvf <study_name>.tgz *.sqlite
2. SFTP the tgz file to the study sqlite archive location on project at CARC (/project/scec_608/cybershake/results/sqlite_studies/<study name>), using hpc-transfer1.
3. Extract the database to /project/scec_608/cybershake/results/sqlite_studies/<study_name> .
tar xzvf sqlite_dumps/<study_name>.tgz <study_name>.sqlite