Storage Policy
Storage policies are available in QuestDB Enterprise only.
A storage policy automates the lifecycle of table partitions. It defines when partitions are converted to Parquet, when they are uploaded to object storage, when local copies are dropped, and when the remote copies are finally reclaimed. Converting a partition to Parquet removes its native files and serves reads directly from the Parquet file. This replaces the need for manual partition management or external scheduling.
The two remote stages, TO REMOTE and DROP REMOTE, drive
cold storage: partitions move to S3, Google
Cloud Storage, Azure Blob Storage, or a filesystem store and stay queryable with
normal SQL.
Requirements
Storage policies require:
- A designated timestamp column
- Partitioning enabled
- QuestDB Enterprise
The two remote stages additionally require:
- Cold storage enabled and configured on every instance
- A WAL-enabled table
Remote stages are rejected on non-WAL tables.
Storage policies do not apply to materialized views at all, local stages included. SET STORAGE POLICY on one is rejected with storage policy is not supported for materialized views. Materialized views use TTL for retention in Enterprise.
How it works
A storage policy consists of up to four TTL settings. Each setting controls a stage in the partition lifecycle:
| Setting | Description |
|---|---|
TO PARQUET | Convert the partition from native binary format to Parquet. The native files are removed and reads are served from the Parquet file |
TO REMOTE | Upload a compact Parquet snapshot of the partition to object storage. The local partition remains writable and is still the serving copy |
DROP LOCAL | Seal the partition as read-only and remove its local data. With TO REMOTE set, reads switch to the remote copy; without it, the data is gone |
DROP REMOTE | Remove the partition from the table and reclaim its remote objects after a grace period |
All settings are optional. Use only the ones relevant to your use case. All TTL
values must be positive; 0 is rejected.
Partition lifecycle
As time passes, each partition progresses through the stages defined by the policy:
TO PARQUET TO REMOTE DROP LOCAL DROP REMOTE
Native ───────┬───────────────────┬───────────────┬─────────────────┬──────
▼ ▼ ▼ ▼
Local Parquet Uploaded, still Sealed, served Removed and
served locally from the store reclaimed
TO PARQUET and TO REMOTE are independent: a partition can be uploaded while
still being served from native local files, and can be converted locally without
ever being uploaded.
DROP LOCAL without TO REMOTE is plain local retention: the partition is
deleted and nothing queryable is left behind. Include TO REMOTE whenever the
partition must stay online after local eviction.
DROP LOCAL is also the point at which a partition becomes read-only. Writes
that arrive for it afterwards are skipped, so set it beyond the longest
late-arrival, replay, and correction window for the table. See
Cold storage.
TTL evaluation
Storage policy TTLs follow the same evaluation rules as TTL. A partition becomes eligible for a lifecycle action when its entire time range falls outside the TTL window:
eligible when: partition_end_time < reference_time - TTL
This rule is applied independently for each stage's TTL. A partition can
be eligible for TO PARQUET long before it is eligible for TO REMOTE,
DROP LOCAL, or DROP REMOTE. Each stage uses its own TTL in the formula
above; the stages share only the reference time and the
ordering constraints.
The reference time is min(wall_clock_time, latest_timestamp) by default —
the same formula used by TTL. The
cairo.ttl.use.wall.clock
setting applies to storage policies as well: setting it to false removes
the wall-clock cap for both TTL and storage policy evaluation. See
TTL § Reference time for the rationale
and the data-loss hazard of disabling the cap.
QuestDB checks storage policies periodically (every 5 minutes by default) and processes eligible partitions automatically.
Storage policy vs TTL
Storage policies replace TTL in QuestDB Enterprise. If you are already familiar with TTL, this comparison is the fastest way in:
| TTL | Storage Policy | |
|---|---|---|
| Availability | Open source | Enterprise only |
| Action | Drops partitions entirely | Graduated lifecycle (convert, then drop) |
| Parquet conversion | No | Yes (automatic local conversion) |
| Granularity | Single retention window | Up to four independent TTL stages |
In QuestDB Enterprise, use storage policies instead of TTL. On a regular table,
ALTER TABLE SET TTL with a non-zero value is rejected, while
CREATE TABLE ... TTL is accepted only for backward compatibility and is
translated into a STORAGE POLICY(DROP LOCAL ...):
-- Instead of:
-- ALTER TABLE trades SET TTL 30 DAYS;
-- Use:
ALTER TABLE trades SET STORAGE POLICY(DROP LOCAL 30d);
A table can carry a TTL only if it was created in QuestDB Open Source and later
upgraded to Enterprise. Clear that legacy TTL with ALTER TABLE SET TTL 0
before setting a storage policy; otherwise SET STORAGE POLICY is rejected with
Cannot set storage policy, please, remove TTL settings. On Enterprise tables
SET TTL 0 is the only accepted SET TTL value; any non-zero value is rejected
with TTL is not supported on Enterprise tables; use a storage policy instead.
Setting a storage policy
At table creation
CREATE TABLE trades (
ts TIMESTAMP,
symbol SYMBOL,
price DOUBLE
) TIMESTAMP(ts) PARTITION BY DAY
STORAGE POLICY(TO PARQUET 3d, DROP LOCAL 1M);
On existing tables
ALTER TABLE trades SET STORAGE POLICY(
TO PARQUET 3 DAYS,
DROP LOCAL 1 MONTH
);
SET STORAGE POLICY replaces the policy in full: every stage you omit is
cleared, not preserved. To keep a stage, restate it in the same statement.
For full syntax details, see ALTER TABLE SET STORAGE POLICY.
TTL duration format
Storage policy TTLs accept the same duration formats as TTL:
| Unit | Long form | Short form |
|---|---|---|
| Hours | 1 HOUR / 2 HOURS | 1h |
| Days | 1 DAY / 3 DAYS | 1d / 3d |
| Weeks | 1 WEEK / 2 WEEKS | 1W / 2W |
| Months | 1 MONTH / 6 MONTHS | 1M / 6M |
| Years | 1 YEAR / 2 YEARS | 1Y / 2Y |
Ordering constraint
The stages form a partial order, not a single chain. A drop stage may not fire before the write stage it depends on:
TO PARQUET <= DROP LOCAL
TO REMOTE <= DROP LOCAL <= DROP REMOTE
TO PARQUET and TO REMOTE are independent: neither has to precede the
other. A TO REMOTE that runs before TO PARQUET uploads a compact Parquet
snapshot while the local partition stays in native format, with reads served
locally from native files until TO PARQUET converts them. All TTL values must
be positive; 0 is rejected.
Disabling and enabling
Temporarily suspend a storage policy without removing it:
ALTER TABLE trades DISABLE STORAGE POLICY;
Re-enable it later:
ALTER TABLE trades ENABLE STORAGE POLICY;
Both ENABLE and DISABLE require a policy to exist on the table; the
statement returns an error otherwise.
Removing a storage policy
To permanently remove a storage policy from a table:
ALTER TABLE trades DROP STORAGE POLICY;
Checking storage policies
Query the storage_policies system view to see all policies and their status:
SELECT * FROM storage_policies;
| table_dir_name | to_parquet | to_remote | drop_local | drop_remote | status | last_updated |
|---|---|---|---|---|---|---|
| trades~12 | 72h | 336h | 1m | 12m | A | 2025-01-15T10:30:00.000000Z |
- TTL values are rendered in two units:
hfor hours andmfor months. Hour-, day-, and week-based durations are stored as hours, so a3 DAYSTTL appears as72hand1 WEEKas168h. Month- and year-based durations are stored as months, so1 MONTHappears as1mand1 YEARas12m. In this viewmmeans months, not minutes; QuestDB's duration shorthand has no unit for minutes - Status
Ameans active;Dmeans disabled (see Disabling and enabling) - An unset stage renders as
0h, not blank
For the full column reference and types, see
storage_policies.
Replication
Storage policy definitions are persisted in WAL-backed system tables, so the policy itself is replicated to every instance in the cluster. Local enforcement runs independently on each instance: Parquet files are produced locally and are not replicated.
This means the primary and its replicas can temporarily disagree on which partitions have been converted to Parquet or dropped, depending on when each node's storage policy check interval last fired. The state converges as each instance processes its own queue.
The remote stages behave differently: one designated instance does the uploading, and partition bytes never travel through the replication stream. See Cold storage roles for how that is divided, and Replication overview for how both kinds of stage behave in a cluster.
Configuration
Storage policy behavior can be tuned in server.conf. Time-based properties
accept values with unit suffixes (e.g., 5m, 1h, 100ms) or raw microsecond
values:
| Property | Default | Description |
|---|---|---|
storage.policy.check.interval | 5m (5 min) | How often QuestDB scans for partitions to process |
storage.policy.retry.interval | 1m (1 min) | Retry interval for failed tasks |
storage.policy.max.reschedule.count | 20 | Maximum retries before abandoning a task |
storage.policy.worker.count | 4 | Number of storage policy worker threads (0 disables the feature) |
storage.policy.worker.affinity | -1 (no affinity) | CPU affinity for each worker thread (comma-separated list) |
storage.policy.worker.sleep.timeout | 100ms | Sleep duration when worker has no tasks |
See Storage policy configuration for the complete list, including the remaining worker-pool tuning properties.
Permissions
Storage policy operations require specific permissions in QuestDB Enterprise:
| Operation | Required permission |
|---|---|
SET STORAGE POLICY | SET STORAGE POLICY |
DROP STORAGE POLICY | REMOVE STORAGE POLICY |
ENABLE STORAGE POLICY | ENABLE STORAGE POLICY |
DISABLE STORAGE POLICY | DISABLE STORAGE POLICY |
Grant permissions using standard RBAC syntax:
GRANT SET STORAGE POLICY ON trades TO analyst;
GRANT REMOVE STORAGE POLICY ON trades TO admin;
End-to-end example
A complete lifecycle on a single table, from creation through verification, modification, inspection of the generated DDL, temporary suspension, and permanent removal:
CREATE TABLE trades (
ts TIMESTAMP,
symbol SYMBOL,
price DOUBLE
) TIMESTAMP(ts) PARTITION BY DAY
STORAGE POLICY(TO PARQUET 3d, DROP LOCAL 1M);
SELECT table_dir_name, to_parquet, drop_local, status
FROM storage_policies
WHERE table_dir_name LIKE 'trades%';
| table_dir_name | to_parquet | drop_local | status |
|---|---|---|---|
| trades~12 | 72h | 1m | A |
ALTER TABLE trades SET STORAGE POLICY(TO PARQUET 1d);
SHOW CREATE TABLE trades;
CREATE TABLE 'trades' (
ts TIMESTAMP,
symbol SYMBOL CAPACITY 256 CACHE,
price DOUBLE
) timestamp(ts) PARTITION BY DAY
STORAGE POLICY(TO PARQUET 1 DAY) WAL;
The DROP LOCAL 1 MONTH stage from step 1 is gone: step 3 restated only
TO PARQUET, and SET STORAGE POLICY replaces the whole policy rather than
merging into it.
ALTER TABLE trades DISABLE STORAGE POLICY;
-- status in storage_policies changes to 'D'
ALTER TABLE trades ENABLE STORAGE POLICY;
ALTER TABLE trades DROP STORAGE POLICY;
-- row disappears from storage_policies
Guidelines
| Use case | Suggested policy | Rationale |
|---|---|---|
| Real-time metrics | TO PARQUET 1d, DROP LOCAL 30d | Keep recent data fast, drop old data automatically |
| Trading data | TO PARQUET 7d | Keep Parquet locally for long-term queries |
| Regulatory history | TO PARQUET 7d, TO REMOTE 14d, DROP LOCAL 90d | Keep years of queryable history without keeping it on local disk |
| IoT telemetry | TO PARQUET 1d, DROP LOCAL 90d | High volume, convert early to save disk before dropping the data |
| Aggregated views | TO PARQUET 30d | Low volume, keep locally in Parquet |
Tips:
- Start with
TO PARQUETto reduce local disk usage while keeping data queryable in Parquet format - Use
DROP LOCALwith care. WithoutTO REMOTEit permanently removes the data; withTO REMOTEit makes the partition read-only for good - Add
DROP REMOTEonly once remote retention is a deliberate decision: it is the only stage that physically deletes data with no local copy left - TTL values should be significantly larger than the partition interval