openstackdb

openstackdb

Typed, read-only Go queries for OpenStack service databases. Extracted from VEXXHOST's OpenStack Database Exporter. The module has no Prometheus or OpenStack API client dependency in its runtime packages.

Layout

Packages follow service boundaries and upstream DB API naming where an operation corresponds. Nova keeps its main (cell) and API databases separate. Octavia exposes repository objects. Placement queries live under placement/objects.

Example

import (
    "context"
    "time"

    "github.com/vexxhost/openstackdb"
    nova "github.com/vexxhost/openstackdb/nova/db/main"
)

func instanceCount(ctx context.Context, databaseURL, projectID string) (int, error) {
    ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
    defer cancel()
    conn, err := openstackdb.ConnectContext(ctx, databaseURL, openstackdb.ConnectionOptions{})
    if err != nil { return 0, err }
    defer conn.Close()

    instances, err := nova.New(conn).InstanceGetAllByFilters(ctx, nova.InstanceFilters{
        ProjectIDs: []string{projectID},
        VMStates: []string{"active", "stopped"},
    })
    return len(instances), err
}

Existing *sql.DB and *sql.Tx values can also be passed to service constructors. Generated query sets support WithTx. Callers own pools, deadlines, region/cell selection, credentials, and authorization. Use database accounts with SELECT access. The library does not discover cells or authorize requests through Keystone.

Filter semantics

Nova instances, Cinder volumes, and Glance images have typed ...GetAllByFilters methods. Fields combine with AND; values in a slice combine with OR. A nil slice means unrestricted, while a non-nil empty slice matches nothing. Filter values are SQL parameters, never SQL fragments. Comparisons use the database collation.

Soft-deleted records are excluded by default. Set Deleted to openstackdb.IncludeDeleted or openstackdb.OnlyDeleted explicitly. DeletedAfter and CreatedBefore must either both be zero or define a nonempty half-open lifetime window: created_at < end AND (deleted_at IS NULL OR deleted_at > start). Times are converted to UTC; these predicates assume UTC database timestamps. Nova InstanceFilters.Archive selects primary tables, shadow tables, or both. WithExtra joins the corresponding instance-extra table for raw flavor JSON. Missing extra rows preserve the instance with a null flavor. Missing requested archive tables are errors; the library never substitutes an incomplete result. Use WithTx with an appropriate transaction isolation level for a consistent primary/shadow snapshot. Rows carry an Archived flag; no deduplication is applied.

IncludeStartBoundary includes rows deleted exactly at the window start. DeletedAtIsNull filters by deletion timestamp independently of the soft-delete flag; combine it with Deleted: openstackdb.IncludeDeleted for timestamp-only selection. Default interval/deletion semantics remain unchanged.

Filtered Nova results use InstanceRecord (including CreatedAt, DeletedAt, Hostname, and optional Flavor). Filtered Glance results use ImageRecord (including DeletedAt). These embed the original exporter row projections. Cinder records preserve nullable volume type IDs for older service schemas.

Metadata and lookup operations

• Nova InstanceExtraGetByInstanceUUID reads flavor JSON, with optional archived access, following upstream instance_extra_get_by_instance_uuid naming.
• Nova InstanceTypeGetAll reads legacy type identities, following historical instance_type_get_all naming. Inactive includes deleted types; Archived selects shadow_instance_types. Modern FlavorGetAll remains in nova/db/api.
• Cinder VolumeTypeGetAll(ctx, VolumeTypeOptions{Inactive: true}) includes deleted types, matching the upstream inactive option. Existing calls remain valid.
• Keystone GetProject returns a project by ID, including disabled projects.

Record sizes and flavor JSON are returned without rounding, unit conversion, or interpretation. Callers own region aggregation and configured default-type fallback.

The richer filtered Nova/Glance result types change explicit result declarations from the initial release; field access remains promoted through the embedded row. The fixed exporter queries and their result projections remain unchanged.

Single-resource methods return sql.ErrNoRows for absent or excluded records. List methods do not promise ordering or pagination; constrain project/resource filters and use deadlines for large deployments.

Upstream correspondence and compatibility

This is a read-only subset inspired by upstream APIs, not a drop-in implementation of their Python APIs. Results are typed projections, not complete ORM objects. See API mapping for naming and retained query semantics.

Exporter-oriented joins and aggregates remain explicit extensions. In particular, Cinder VolumeGetAllWithAttachments can return multiple rows per volume; VolumeGetAll and VolumeGetAllByFilters return one row per volume. Manila's ShareGetAllWithInstances and Octavia's GetAllWithVIP similarly describe joins. No resource mutation, schema migration, or OpenStack request-context behavior is implemented. Query compatibility depends on the deployed service schema.

schema.Files embeds the minimal schema/index fixtures used by integration tests. They are test inputs, not production migrations or a claim of support for every OpenStack release. Some inherited queries require indexes present in the fixtures.

Development

Requires Go 1.25.5 or later. SQL sources live alongside their generated Go packages. Do not edit queries.sql.go, models.go, or generated db.go files by hand.

go install github.com/sqlc-dev/sqlc/cmd/sqlc@v1.30.0
sqlc generate
go test -race ./...
go vet ./...
# Requires Docker; runs every query against MariaDB and seeded filter scenarios.
go test -race -tags integration -timeout 15m ./...

Exporter collector tests continue to verify metric output in the exporter repository. This module owns connection, predicate, and database query tests.

License

Apache-2.0. See LICENSE and NOTICE.