HSn0918 opened a new issue, #70778:
URL: https://github.com/apache/doris/issues/70778
## Version
Apache Doris `doris-4.1.5-release-917b20be838` on both FE and BE. Reproduced
with one FE and one BE in `apache/doris:all-in-one-4.1.5` on Linux ARM64.
## What's Wrong?
We first saw this on a production cluster with three BEs while investigating
slow PCA queries and repeated full refreshes of two async materialized views.
Data imports were running during the investigation, and a refresh had hit a
BE's spill limit. When we opened the views in the web UI to inspect their
columns, the field list repeatedly failed to load.
On November 20, 2026, the production FE logged 19 occurrences of the same
exception between 09:39:15 and 10:42:00. Each stack trace included
`MetaInfoAction.getTableSchema`. Over SQL, `DESC` returned 80 columns for one
view and 17 for the other, and `SELECT ... LIMIT 1` succeeded on both. Each
view's actual base index ID was one greater than its table ID.
We then reproduced the field-list failure on a local 4.1.5 cluster with one
row, after both test views had finished refreshing. The FE schema endpoint
returned HTTP 200 with `"code":500`, `"msg":"Internal Error"`, and the
following error in `data`.
```text
Cannot invoke "org.apache.doris.catalog.MaterializedIndexMeta.getSchema()"
because the return value of "java.util.Map.get(Object)" is null
```
## What You Expected?
The endpoint should return the view's base schema. For the example below,
its field names should match the two columns returned by `DESC mv_detail`.
## How to Reproduce?
On a cluster with one FE and one BE, run:
```sql
CREATE DATABASE schema_repro;
USE schema_repro;
CREATE TABLE base_rows (k INT, v INT)
DUPLICATE KEY(k)
DISTRIBUTED BY HASH(k) BUCKETS 1
PROPERTIES ("replication_num" = "1");
INSERT INTO base_rows VALUES (1, 7);
CREATE MATERIALIZED VIEW mv_detail
BUILD IMMEDIATE REFRESH COMPLETE ON MANUAL
DISTRIBUTED BY RANDOM BUCKETS 1
PROPERTIES ("replication_num" = "1")
AS SELECT k, v FROM base_rows;
```
Wait for `RefreshState` to become `SUCCESS`, then describe and query the
view:
```sql
SELECT Name, State, RefreshState FROM mv_infos('database'='schema_repro');
DESC mv_detail;
SELECT k, v FROM mv_detail;
```
From inside the local all-in-one container, call the same FE endpoint for
both objects:
```sh
tables=http://127.0.0.1:8030/api/meta/namespaces/default_cluster/databases/schema_repro/tables
curl -sS -i -u root: "$tables/base_rows/schema"
curl -sS -i -u root: "$tables/mv_detail/schema"
```
## Anything Else?
The base table returned `code: 0` and fields `k, v` on three consecutive
requests. The projection view above and a second async view defined with
`SELECT k, SUM(v) AS total FROM base_rows GROUP BY k` each returned the same
error on all three requests. Both views reported `NORMAL / SUCCESS`, and both
returned one row with values `(1, 7)` over SQL.
In `MetaInfoAction.getTableSchema`, the FE reads `getBaseIndexId()` only for
objects of type `OLAP`. An async materialized view has type
`MATERIALIZED_VIEW`, so the handler builds its `index_schema` proc path with
`tableId - 1`. In the local run, `mv_detail` had table ID `1795143523871`. The
computed index ID was `1795143523870`, while `SHOW PROC` listed its only index
as `1795143523872`.
Relevant FE stack frames:
```text
at org.apache.doris.catalog.OlapTable.getSchemaByIndexId(OlapTable.java:1143)
at org.apache.doris.catalog.OlapTable.getSchemaByIndexId(OlapTable.java:1139)
at
org.apache.doris.common.proc.IndexInfoProcDir.lookup(IndexInfoProcDir.java:127)
at org.apache.doris.common.proc.ProcService.open(ProcService.java:143)
at
org.apache.doris.httpv2.rest.MetaInfoAction.generateSchema(MetaInfoAction.java:289)
at
org.apache.doris.httpv2.rest.MetaInfoAction.generateResult(MetaInfoAction.java:285)
at
org.apache.doris.httpv2.rest.MetaInfoAction.getTableSchema(MetaInfoAction.java:260)
```
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]