[GEODE-9392] A gfsh query returning a Struct containing a PdxInstance behaves differently than one returning just the PdxInstance in some cases

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: gfsh
• Assignee: Unassigned
• Reporter: Barrett Oglesby
• Created: 21/Jun/21 23:30
• Updated: 14/Jul/21 23:23

Description

This is true when the PdxInstance contains a data type that is not supported by PdxToJSON (like Date or Character).

If objects like this are stored as PdxInstances:

public class Position {
  private String id;
  private Date tradeDate;
  private Character type;
  ...
}

A query like this is successful:

Executing - query --query='select * from /positions'

Result : true
Limit  : 100
Rows   : 10

  tradeDate   | id | type
------------- | -- | ----
1624316618413 | 3  | "a"
1624316618324 | 0  | "a"
1624316618418 | 5  | "a"
1624316618421 | 6  | "a"
1624316618407 | 1  | "a"
1624316618426 | 8  | "a"
1624316618428 | 9  | "a"
1624316618415 | 4  | "a"
1624316618423 | 7  | "a"
1624316618410 | 2  | "a"

But a query like this is not:

Executing - query --query="select key,value from /positions.entries where value.id = '0'"

Result  : false
Message : Could not create JSON document from PdxInstance

It fails with this exception in the server:

org.apache.geode.pdx.JSONFormatterException: Could not create JSON document from PdxInstance
	at org.apache.geode.pdx.JSONFormatter.fromPdxInstance(JSONFormatter.java:241)
	at org.apache.geode.pdx.JSONFormatter.toJSON(JSONFormatter.java:226)
	at org.apache.geode.management.internal.cli.domain.DataCommandResult$SelectResultRow.valueToJson(DataCommandResult.java:732)
	at org.apache.geode.management.internal.cli.domain.DataCommandResult$SelectResultRow.resolveStructToColumns(DataCommandResult.java:717)
	at org.apache.geode.management.internal.cli.domain.DataCommandResult$SelectResultRow.resolveObjectToColumns(DataCommandResult.java:692)
	at org.apache.geode.management.internal.cli.domain.DataCommandResult$SelectResultRow.createColumnValues(DataCommandResult.java:680)
	at org.apache.geode.management.internal.cli.domain.DataCommandResult$SelectResultRow.<init>(DataCommandResult.java:663)
	at org.apache.geode.management.internal.cli.functions.DataCommandFunction.createSelectResultRow(DataCommandFunction.java:270)
	at org.apache.geode.management.internal.cli.functions.DataCommandFunction.select_SelectResults(DataCommandFunction.java:256)
	at org.apache.geode.management.internal.cli.functions.DataCommandFunction.select(DataCommandFunction.java:224)
	at org.apache.geode.management.internal.cli.functions.DataCommandFunction.select(DataCommandFunction.java:177)
	at org.apache.geode.management.internal.cli.functions.DataCommandFunction.execute(DataCommandFunction.java:126)
Caused by: java.lang.IllegalStateException: PdxInstance returns unknwon pdxfield tradeDate for type Mon Jun 21 16:03:38 PDT 2021
	at org.apache.geode.pdx.internal.json.PdxToJSON.writeValue(PdxToJSON.java:148)
	at org.apache.geode.pdx.internal.json.PdxToJSON.getJSONString(PdxToJSON.java:185)
	at org.apache.geode.pdx.internal.json.PdxToJSON.getJSON(PdxToJSON.java:61)
	at org.apache.geode.pdx.JSONFormatter.fromPdxInstance(JSONFormatter.java:239)

Its because of the difference in processing a PdxInstance (first query) and a Struct (second query) in resolveObjectToColumns:

private void resolveObjectToColumns(Map<String, String> columnData, Object value) {
  if (value instanceof PdxInstance) {
    resolvePdxToColumns(columnData, (PdxInstance) value);
  } else if (value instanceof Struct) {
    resolveStructToColumns(columnData, (StructImpl) value);
  }
  ...
}

They both end up in SelectResultRow.valueToJson:

private String valueToJson(Object value) {
  ...
  if (value instanceof String) {
    return (String) value;
  }

  if (value instanceof PdxInstance) {
    return JSONFormatter.toJSON((PdxInstance) value);
  }

  ObjectMapper mapper = new ObjectMapper();
  try {
    return mapper.writeValueAsString(value);
  } catch (JsonProcessingException jex) {
    return jex.getMessage();
  }
}

In the PdxInstance case, the fields are passed in individually and handled by the first condition (String) and the ObjectMapper (Date, Character):

SelectResultRow.resolveObjectToColumns value=PDX[13681235,Position]{id=3, tradeDate=Mon Jun 21 16:03:38 PDT 2021, type=a}; valueClass=class org.apache.geode.pdx.internal.PdxInstanceImpl
SelectResultRow.valueToJson value=Mon Jun 21 16:03:38 PDT 2021; valueClass=class java.util.Date
SelectResultRow.valueToJson value=3; valueClass=class java.lang.String
SelectResultRow.valueToJson value=a; valueClass=class java.lang.Character

In the Struct case, the value passed in is a Pdxinstance which is handled by the JSONFormatter:

SelectResultRow.resolveObjectToColumns value=struct(key:0,value:PDX[13681235,Position]{id=0, tradeDate=Mon Jun 21 16:03:38 PDT 2021, type=a}); valueClass=class org.apache.geode.cache.query.internal.StructImpl
SelectResultRow.valueToJson value=0; valueClass=class java.lang.String
SelectResultRow.valueToJson value=PDX[13681235,Position]{id=0, tradeDate=Mon Jun 21 16:03:38 PDT 2021, type=a}; valueClass=class org.apache.geode.pdx.internal.PdxInstanceImpl

JSONFormatter delegates to PdxToJSON which doesn't understand Characters or Dates and throws the IllegalStateException.

If I change the last else clause in PdxToJSON.writeValue so that it uses ObjectMapper like SelectResultRow.valueToJson does like:

} else {
  ObjectMapper mapper = new ObjectMapper();
  try {
    jg.writeString(mapper.writeValueAsString(value));
  } catch (JsonProcessingException jex) {
    jg.writeString(jex.getMessage());
  }
}

The query is successful:

Executing - query --query="select key,value from /positions.entries where value.id = '0'"

Result : true
Limit  : 100
Rows   : 1

key | value
--- | -----------------------------------------------------
0   | {"tradeDate":"1624317757557","id":"0","type":"\"a\""}

I'm not sure if this is a valid fix, but it will handle a lot more cases than PdxToJSON.writeValue does now.