Database Migration

Concept

Migration tool creates source_hash and target_hash table for all tables and store a primary key(s) and the MD5 hash of the rest of the columns. Hash is computed for all rows, on source and target as well.

Prerequisites

• Supported Codebeamer versions:
  • 22.04 or higher SPs
  • 22.10-SP7 or higher SPs
  • 2.0.0.0 or higher SPs
  • 2.1.0.0 or higher SPs
• Linux machine with Installed docker.
• Oracle 19.x or PostgreSQL 12.x database.
  • Source database must be available only for the migration tool.
  • Empty schema must be created in the target database.
• Artifact in Database feature is not used.
• Enough free disk for new source hash table on the source and target database. It is recommended to have half the allocation disk empty.

Limitation

Temp tables will not be dropped after migration in the source database. Temp tables end with _temp .

How to Create PostgreSQL

Following configuration should be used for database creation:

LC_COLLATE='en_US.UTF-8' LC_CTYPE='en_US.UTF-8' ENCODING 'UTF8'

How to Create Oracle

Following configuration should be used for database creation:

GRANT CREATE SESSION, CREATE TYPE, CREATE TABLE, CREATE CLUSTER, CREATE TRIGGER, CREATE OPERATOR, CREATE SEQUENCE, CREATE INDEXTYPE, CREATE PROCEDURE, CREATE VIEW
GRANT EXECUTE ON CTXSYS.CTX_DDL
GRANT EXECUTE ON DBMS_LOB
GRANT EXECUTE ON DBMS_CRYPTO

DBMS_CRYPTO grant can be removed after the migration.

Migration Stages

Pre-Stage - Steps are running on the source database

1. Create temporally tables with PKs for history_search_update, initial_version, locks, session_user_licenses, task_reference_tag, user_key, reference_search_history, task_field_value, field_value_search_history, project_activity_log, computed_field_lookup, object_access_permission, object_notification, object_reference, object_approval_history, object_reference_filter, task_field_status_specific
2. Clean up invalid data.
3. Generate source hash.
4. Assert source hash tables generation was successful.

Java-Migration-Stage

1. Migrate data from source to target

Post-Stage - Steps are running on target database

1. Create target database.
2. Create sequences.
3. Create tables.
4. Create source hash tables.
5. Set version number.
6. Generate target hash.
7. Assert target hash.
8. Validate source and target hash.
9. Drop source and target hash tables.
10. Add View and Procedure.
11. Remove invalid data.
12. Add indexes.
13. Add PKs.
14. Add constraints and FKs.

How to Run the Migration Tool

Configuration

Create a configuration file using the following template.

config.conf

# Number of threads is used for running hash computation script. Use the number of CPUs of source database.
db.parallel.sql_job_number=4

# Number of threads is used for running PK creation script. Use the number of CPUs of source database.(max 11)
db.parallel.pk_job_number=2

db.source.database.type=<mysql|postgresql|oracle>
db.target.database.type=<postgresql|oracle>

db.source.host=<database host>
db.source.port=<database port>
db.source.username=<database username>
db.source.password=<database password>
db.source.database=<database name>

db.target.host=<database host>
db.target.port=<database port>
db.target.username=<database username>
db.target.password=<database password>
db.target.database=<database name>

# default is true
migrator.run.pre_stage=<true|false>

# default is true
migrator.run.java_stage=<true|false>

# default is true
migrator.run.post_stage=<true|false>

# Run hash computation and validation or not
migrator.verify_rows=<true|false>

# It depends on your memory of the migration server and database server
migrator.fetch_size=3000

# It depends on your memory of the migration server and database server
migrator.batch_size=3000

# Number of maximum threads is used for running migration script. Use the number of CPUs of your migration machine
migrator.worker.threads=12

# Number is threads is used for processing one table, smaller or same as the total.sections configuration
migrator.table.parallel=12

# Number of sections that table is cut of
migrator.total.sections=24

# Drop or keep the hash tables, it is helpful for investigating migration issues, by default it is true
migrator.drop_hash_tables=<true|false>

# See the related section below, by default it is false
migrator.drop_document_cache=<true|false>

# See the related section below, by default it is true
migrator.update_boolean_values=<true|false>

# See the related section below, by default it is true
migrator.remove_000_characters=<true|false>

# Version of the source codebeamer
cb.major.version=22.10

# Version of the source codebeamer
cb.minor.version=-SP7

How to Download the Migrator Tool

Docker Images

• intland/database-migration-tool:22.04-1.2
• intland/database-migration-tool:22.10-1.2
• intland/database-migration-tool:2.0-1.2
• intland/database-migration-tool:2.1-1.2

How to Pull Docker Image

1. docker login -u intlandsnapshot -p 89bc690e-bf19-4014-885b-7cd4c6355ca2
2. docker pull <docker image>
3. docker logout

How to Start the Migration Tool

docker run -v <path of config.conf>:/conf/config -v <path of directory>:/conf/logs -i <docker image, see above> /rootfs/run-migration.sh

Database Configuration

PostgreSQL Tuning

synchronous_commit = off

See: https://www.postgresql.org/docs/current/wal-async-commit.html#:~:text=Asynchronous%20commit%20is%20an%20option,is%20an%20acceptable%20trade%2Doff.

Oracle

nls_length_semantics = CHAR

Execution Times of Test Runs

We used a ~200G database running on AWS.

Details:

• Source RDS is a db.m5.2xlarge (8 vCPU / 32G / 4000 IOPS) server
• Target RDS are a db.m5.2xlarge (8 vCPU / 32G / 8000 IOPS) server
• Migration server is a m5.2xlarge (8 vCPU / 32G / 2000 IOPS) server
• Network is up to 10 Gbps

Execution Time for MySQL -> Oracle

With Verification

• Add PKs using 2 threads - 2h 30m
• Source hash generation using 4 threads - 1h 10m
• Migrate Source to Target - 5h 30m
• Target hash generation using 4 threads - 22m
• Validate source and target hash - 8m
• Drop temporally Pks using 4 threads - 1h
• Remove invalid data - 7m
• Create indexes using 4 threads - 7m
• Add PKs using 4 threads - 2m

Without Verification

• Add temporally PKs using 2 threads - 2h 30m
• Migrate Source to Target - 1h 50m
• Drop temporally Pks using 4 threads - 1h
• Remove invalid data - 7m
• Create indexes using 4 threads - 7m
• Add PKs using 4 threads - 2m

Execution Time for MySQL -> PostgreSQL

With Verification

• Add temporally PKs using 2 threads - 2h 30m
• Source hash generation using 4 threads - 1h 10m
• Migrate Source to Target - 5h
• Drop temporally Pks using 4 threads - 1m
• Target hash generation using 4 threads - 10m
• Validate source and target hash - 21m
• Remove invalid data - 14m
• Create indexes using 4 threads - 12m
• Add PKs using 4 threads - 2m

Without Verification

• Add temporally PKs using 2 threads - 2h 30m
• Migrate Source to Target - 1h 40m
• Drop temporally Pks using 4 threads - 1m
• Remove invalid data - 14m
• Create indexes using 4 threads - 12m
• Add PKs using 4 threads - 2m

Recommendation

Half of the migration time is the verification step, in order to speed up the migration.

Dry-run migration with verification and validation:

migrator.verify_rows=true
migrator.drop_hash_tables=false

Real migration with verification and validation:

migrator.verify_rows=true
migrator.drop_hash_tables=true

Real migration without verification and validation:

migrator.verify_rows=false
migrator.drop_hash_tables=false

Remove Data of document_cache

document_cache, document_cache_data, and document_cache_data_blobs contain data for faster re-index of the database. These tables can be truncated before the migration:

SET foreign_key_checks = 0;
TRUNCATE TABLE document_cache;
TRUNCATE TABLE document_cache_data;
TRUNCATE TABLE document_cache_data_blobs;
SET foreign_key_checks = 1;

Data Inconsistency

If migrator_update_boolean_values is set to true, on MySQL the following SQL scripts will be executed in order to fix data inconsistency:

UPDATE object_notification_temp SET only_members = 0 WHERE only_members IS NULL;
UPDATE object_notification_temp SET deleted = 0 WHERE deleted IS NULL;
UPDATE task_type SET service_desk = 0 WHERE service_desk IS NULL;
UPDATE task_type SET template = 0 WHERE template IS NULL;

If migrator.remove_000_characters is set to true, on MySQL the following SQL scripts will be executed in order to fix data inconsistency:

UPDATE task_field_history SET new_value = replace(new_value, 0x00, "") WHERE new_value LIKE CONCAT("%", CHAR(0x00 using utf8), "%");
UPDATE task_field_history SET old_value = replace(old_value, 0x00, "") WHERE old_value LIKE CONCAT("%", CHAR(0x00 using utf8), "%");
UPDATE task_field_value SET field_value = replace(field_value, 0x00, "") WHERE field_value LIKE CONCAT("%", CHAR(0x00 using utf8), "%");