🔗 Ligoj Provisioning plugin

Ligoj Provisioning plugin Provides the following features :

• Find the best instance from the given requirement : CPU, RAM, OS and price type
• Compute the total cost of storage and compute for a set of VM

Supported requirements

Name	Note
Term	A contract, defining the constraints: reservation, conversion options and sometimes a location
Location	A geographical place (variable GPS coordinates precision). May be a country, or sometimes a city
Processor	The underlying physical processor. The vendor can be used instead of the full product code.
Architecture	The underlying physical processor architecture, such as x86 or ARM.
Physical	A boolean constraint to expect a physical (bar metal) instance instead of a virtual one
Tenancy	Shared or dedicated, to handle the noisy neighborhood issue
OS	Operating system. Currently, only the type name is supported: Windows, Linux, RHEL,...
Software	The pre-installed software
Engine	Database engine name. Can be MySQL, Oracle, .. or even custom one like Aurora. Depends on the availability in the catalog.
Edition	Database edition valid for a specific engine. For sample: Oracle Standard Edition 1
License	BYOL or included mode. Depends on the availability in the catalog.
Optimized	Storage expected optimization: durability, IOPS, throughput
CPU/RAM	Expected vCPU/RAM(MiB) to match. Note that some providers support custom (elastic) settings for these values. Ligoj handles this.
Ephemeral	A boolean to accept a shutdown of a instance. Would be plugged to Spot/Batch like services
Usage	Utilization profile including up-time, commitment, conversion, reservation and forecasted start.
Size	Storage size in GiB. The required value is checked against provider limits.
Latency	IO latency access rating from WORST to BEST
Optimized	Optimization profile: durability, throughput and IOPS
Generation	(p1TypeOnly) Instead of searching a cheapest instance, find the cheapest instances among the last generations SKUs

Covered requirements per resource

Resource	Criteria
Instance	Tenancy, processor, physical (metal), architecture, ephemeral, %usage, location, term, cpu, ram, OS, software, p1TypeOnly
Database	Tenancy, processor, physical (metal), architecture, ephemeral, %usage, location, term, cpu, ram, engine, edition, p1TypeOnly
Storage	Location, size, latency, location, optimization
Support	Phone, mail, chat, API, seats

Covered service per provider

Provider	Plugin	Covered services
AWS	plugin-prov-aws	EC2 (Savings Plan, RI, Spot), RDS, S3 (Glacier, IA,...), EFS, Support, EBS,
Azure	plugin-prov-azure	VM, Database, Disk (object, file, block),
Digital Ocean	plugin-prov-digitalocean	VM, Database, Disk
OVH	plugin-prov-ovh	VM, Database, Disk
OutScale	plugin-prov-outscale	VM, Database, Disk

Related plugins:

• plugin-prov-aws
• plugin-prov-azure
• plugin-prov-digitalocean
• plugin-prov-outscale
• plugin-prov-ovh

Custom configuration: service:prov:use-parallel = 0/1

Catalog import and the database

Provider catalog imports that opt in (currently plugin-prov-azure) run inside a single database transaction: prices are accumulated in the JPA persistence context, flushed and cleared by chunks of 1000 prices, and written with a session-scoped JDBC batch size of 100 (AbstractImportCatalogResource: initJdbcBatch(), flushChunk(), flushAndClear()). No global Hibernate configuration is required. On failure, the whole import rolls back and the previous catalog is preserved.

Database recommendations for large imports (millions of prices):

• idle_in_transaction_session_timeout (PostgreSQL) must be 0 (no limit, the default) or comfortably above a few minutes: the import transaction sits idle while the provider catalog files (tens of MB) are downloaded and parsed. The ligoj-cli managed ligoj-db pod pins -c idle_in_transaction_session_timeout=0 explicitly since 2026-07, so a base-image or ALTER SYSTEM change cannot silently kill a running import.
• Add reWriteBatchedInserts=true to the PostgreSQL JDBC URL (jdbc.url) so the JDBC batches are rewritten as multi-row INSERT statements.