collado-mike commented on code in PR #1378:
URL: https://github.com/apache/polaris/pull/1378#discussion_r2053108958
##########
service/common/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalog.java:
##########
@@ -1328,6 +1346,41 @@ public void doCommit(TableMetadata base, TableMetadata
metadata) {
String newLocation = writeNewMetadataIfRequired(base == null, metadata);
String oldLocation = base == null ? null : base.metadataFileLocation();
+ // TODO: we should not need to do this hack, but there's no other way to
modify
+ // currentMetadata / currentMetadataLocation
Review Comment:
Ugh. I see the problem - it is here:
https://github.com/apache/iceberg/blob/main/core/src/main/java/org/apache/iceberg/TableMetadata.java#L985
. The `TableMetadata` builder copies the `changes` field from the original, so
when we set the `metadataLocation`, it looks like we're trying to set it on an
object that has changes already 🙄
We _can_ do this without reflection if we roundtrip to JSON. Something like
this diff does work. It does require a roundtrip through Jackson, so there's a
performance cost 😢. Worth considering whether we'd rather pay the perf cost or
deal with reflection in the code. The perf cost would scale with the size of
the `TableMetadata`, though for larger metadata files, we're probably not
loading the `TableMetadata` in memory at all.
```diff
diff --git
a/service/common/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalog.java
b/service/common/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalog.java
index 2e8fadab..8d77fa84 100644
---
a/service/common/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalog.java
+++
b/service/common/src/main/java/org/apache/polaris/service/catalog/iceberg/IcebergCatalog.java
@@ -1242,6 +1242,9 @@ public class IcebergCatalog extends
BaseMetastoreViewCatalog
SHOULD_RETRY_REFRESH_PREDICATE,
getMaxMetadataRefreshRetries(),
metadataLocation -> {
+ if (tblMetadata.containsKey(metadataLocation)) {
+ return tblMetadata.get(metadataLocation);
+ }
String latestLocationDir =
latestLocation.substring(0,
latestLocation.lastIndexOf('/'));
// TODO: Once we have the "current" table properties pulled
into the resolvedEntity
@@ -1259,6 +1262,8 @@ public class IcebergCatalog extends
BaseMetastoreViewCatalog
}
}
+ Map<String, TableMetadata> tblMetadata = new HashMap<>();
+
@Override
public void doCommit(TableMetadata base, TableMetadata metadata) {
LOGGER.debug(
@@ -1400,9 +1405,9 @@ public class IcebergCatalog extends
BaseMetastoreViewCatalog
// TODO: we should not need to do this hack, but there's no other way
to modify
// currentMetadata / currentMetadataLocation
if (updateMetadataOnCommit) {
- TableOperationsReflectionUtil reflectionUtil =
TableOperationsReflectionUtil.getInstance();
- reflectionUtil.setMetadataFileLocation(metadata, newLocation);
- reflectionUtil.setCurrentMetadata(this, metadata, newLocation);
+ TableMetadata updated =
+ TableMetadataParser.fromJson(newLocation,
TableMetadataParser.toJson(metadata));
+ tblMetadata.put(newLocation, updated);
}
if (null == existingLocation) {
createTableLike(tableIdentifier, entity);
```
--
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]
