shaoyu-li commented on code in PR #13058:
URL: https://github.com/apache/gravitino/pull/13058#discussion_r3985842897
##########
catalogs/catalog-lakehouse-generic/src/main/java/org/apache/gravitino/catalog/lakehouse/generic/GenericCatalogOperations.java:
##########
@@ -273,57 +299,80 @@ public Table alterTable(NameIdentifier ident,
TableChange... changes)
@Override
public boolean purgeTable(NameIdentifier ident) {
- boolean purged = tableOps(ident).purgeTable(ident);
- tableFormatCache.invalidate(ident);
- return purged;
+ return dropOrPurgeTable(ident, true /* purge */);
}
@Override
public boolean dropTable(NameIdentifier ident) throws
UnsupportedOperationException {
- boolean dropped = tableOps(ident).dropTable(ident);
- tableFormatCache.invalidate(ident);
- return dropped;
+ return dropOrPurgeTable(ident, false /* purge */);
}
- private String calculateTableLocation(
- Schema schema, NameIdentifier tableIdent, Map<String, String>
tableProperties) {
- String tableLocation =
- (String)
- propertiesMetadata
- .tablePropertiesMetadata()
- .getOrDefault(tableProperties, Table.PROPERTY_LOCATION);
- if (StringUtils.isNotBlank(tableLocation)) {
- return ensureTrailingSlash(tableLocation);
+ /**
+ * Drops or purges a table, and hands its location back to the {@link
TableLocationProvider}
+ * afterwards.
+ *
+ * <p>The table properties are read before the removal, because they carry
the location the
+ * provider has to hand back, and the unprovisioning itself happens after
the removal so that a
+ * provider never reclaims the storage of a table that is still there. A
provider failing to
+ * unprovision is logged at WARN rather than propagated: the table is
already gone at that point,
+ * so failing the request would report a drop that did in fact happen as
unsuccessful and invite a
+ * retry that cannot undo anything.
+ *
+ * @param ident the identifier of the table to drop
+ * @param purge whether to purge the table instead of dropping it
+ * @return true if the table was dropped, false if it did not exist
+ */
+ private boolean dropOrPurgeTable(NameIdentifier ident, boolean purge) {
+ Map<String, String> tableProperties;
+ try {
+ tableProperties = store.get(ident, TABLE,
TableEntity.class).properties();
+ } catch (NoSuchEntityException e) {
+ return false;
+ } catch (IOException e) {
+ throw new RuntimeException(
+ String.format("Failed to load table %s before dropping it", ident),
e);
}
- String schemaLocation =
- schema.properties() == null ? null :
schema.properties().get(Schema.PROPERTY_LOCATION);
-
- // If we do not set location in table properties, and schema location is
set, use schema
- // location as the base path.
- if (StringUtils.isNotBlank(schemaLocation)) {
- return ensureTrailingSlash(schemaLocation) + tableIdent.name() + SLASH;
- }
+ // Built entirely before the drop, so that a store read failing here fails
the request while
+ // the table is still there, rather than after it is gone where it could
only be reported as a
+ // provider failure it is not.
+ TableLocationContext context =
+ TableLocationContext.builder()
+ .withTableIdentifier(ident)
+ .withTableProperties(tableProperties)
+
.withSchema(loadSchema(NameIdentifier.of(ident.namespace().levels())))
Review Comment:
One of your three reads I should have answered and did not: the third one,
`LanceTableOperations.dropTable` calling `super.loadTable`. That read is
inside the format, below
this SPI, and it is there because the format needs the table's own
properties to decide what to do
with the data. Removing it would mean widening `TableOperations.dropTable`
to carry properties down
-- a change to a shared interface, for a saving on a path that is already
writing to the entity
store. I would rather not fold that into this PR, but say so if you
disagree; it is a small change,
just a wider one.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
