workspace.preferences can trail a change by about a second. A Provider
connected a moment ago can take that long to appear. To query this pool, you
need the Data Access, Member, Member with All Data Access or Workspace
Administrator role.
Secrets, API tokens, sessions, and credential material are never in this
pool.
Tables
IDs are served in base62 form, such as
YQJpYwUwvbaLOwTUr4thA. An ID that you
type into a comparison matches in base62 or hyphenated form.
Workspace Pool
lists every column of every table.
List active Providers
Every connection is a row, disabled ones included. “Active” is a filter that you write:IS DISTINCT FROM. The SQL parser reads
everything after IS DISTINCT FROM as its right side, so
connection_status IS DISTINCT FROM 'disabled' AND setup_finished compares
connection_status with 'disabled' AND setup_finished, and the query is
refused with QL-PL-019.
Add AND setup_finished to leave out a setup that was started and never
finished. Once its status is recomputed, a pull connection that never saved a
credential reads config_invalid and counts as set up, so it shows as needing
attention (PARABLE-9399).
connector_slug is the {connector} of providers.{connector}.{table} in
catalog form: lowercase, with each run of other characters written as one
_. The claude-otel Provider is providers.claude_otel.
Read Provider settings
Each connected Provider gets its settings as tables, one row per connection.tenant_connector_id names the connection and joins workspace.providers.id.
Column names follow the Provider’s settings, lowercased. When the Provider’s
settings change shape, the table changes with them.
For Google Workspace, workspace.google_ingestion_config has domain and
adminemail. workspace.google_authentication_config has non-secret fields
such as clientemail and projectid. A field that the Provider marks secret,
such as a private key, is never a column. A list of objects in the settings
becomes a child table named {table}__{field}, joined on parent_pk and
ordinal.
workspace.parent_data_filters has one row per allowed value, with
connector_slug, parent_tap, field, operator, value, and threshold.
Which Google users is the sync limited to?
Members and roles
Roles are defined once for the platform, not for each Workspace.workspace.roles lists the roles that at least one member holds. Join on the
role id, never on its name, because an administrator can rename a role.
Who holds each role:
u.status = 'ACTIVE' to see who can act today.
A permission is a dotted path, and a grant covers every path below it. A role
granted tenant can do everything under tenant., such as
tenant.users.manage. workspace.role_permissions lists direct grants only.
It does not expand the hierarchy, and it does not include what a role inherits
from its parent role. Which roles cover one permission:
Plot run history and spend
workspace.query_runs has one row for each run of a Plot. spent_cost is
what its ponder() and embed() calls cost, in US cents. spent_calls is
how many calls it made. What each Parable spent in the last 30 days:
workspace.parables lists only the Parables that the Workspace takes part
in. This join therefore leaves out the runs of a Parable that a person built
and did not share. workspace.query_runs itself lists every run in the
Workspace. A run’s SQL is never served.
Keep a scan small
One scan of aworkspace.* table reads at most 50,000 rows. A scan that
would read more fails with RESOURCE_EXHAUSTED. workspace.query_runs keeps
every run, so a large Workspace can pass that limit.
Parable applies these filters in the database, before it counts rows:
=on text, integer, boolean, ID, and timestamp columns.INon text, integer, and ID columns, with up to 1,000 values.IS NULLandIS NOT NULLon every column.<>,IS DISTINCT FROM, andNOT INon ID columns.<,<=,>, and>=on timestamp columns.
spent_cost > 100
or status LIKE 'FAIL%', apply after the read and do not make it smaller.
Combine them with a filter from the list.
Which parents a sync could not read
A sync of a child stream reads each parent record, such as each repository of a GitHub organization. When the Provider refuses one parent, the sync still completes.workspace.provider_refusals has one row for each refused parent
of each sync attempt. A retry adds its own rows, so a refusal that a later
retry cleared still counts:
GROUP BY refusal.http_status.
Which days of a stream landed
workspace.stream_coverage has one row for each stream and each of the last
7 full UTC days. A day is covered when the stream’s loads span it from
midnight to midnight UTC. gap_cause says why a day is not covered:
no_seal (nothing loaded the day) or partial_seal (loads touched the day
but did not span it). reached_pool says whether the day’s data is in the
Provider pool:
Join live metadata to Provider data
workspace.users.email and Provider columns with the Contact.Email semantic
type hold compatible values. Members and their Google Workspace directory
rows: