[SOLR-15411] "failOnVersionConflicts" doesn't get checked in getUpdatedDocument when _version_==1

• Type: Bug
• Status: Patch Available
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 8.5.1
• Component/s: UpdateRequestProcessors
• Assignee: Unassigned
• Reporter: xuanyu huang
• Created: 18/May/21 11:39
• Updated: 26/Feb/24 00:00

Description

I'm using solrj for Java to update 4 docs:

 

UpdateRequest updateRequest = new UpdateRequest();    
updateRequest.setAction( UpdateRequest.ACTION.COMMIT, false, false);

for (Map<String, Object> map : maps) {      
     if (map.containsKey("record_uuid")) {        
          SolrInputDocument doc = new SolrInputDocument();
          doc.addField("assertionUserId", new HashMap<String, Object>(){{ put("set", map.getOrDefault("assertionUserId", null)); }});
          // other doc.addField() 
          logger.debug("Added solr doc for record: " + doc.get("id"));                  
          updateRequest.add(doc);      
      }    
}

// update only when there are docs to update    
if (updateRequest.getDocuments() != null) {          
   updateRequest.setParam("_version_", "1");         
   updateRequest.setParam("failOnVersionConflicts", "false");         
   updateRequest.process(solrClient);      
 
}
 

There are 4 docs added into the updateRequest and the 2nd doc has an invalid id. 
 
See that I've set the _version_ to 1 and failOnVersionConflicts to false
 
When I ran the program I could see this log
 
 

- Added solr doc for record: id=2c9b8671-a55a-40ab-940f-06f9cf987880
- Added solr doc for record: id=c0ee1a86-1df6-40b2-950c-bdde40b1c46e_invalid
- Added solr doc for record: id=af56ce03-e664-421a-85ac-fbb839bbb140
- Added solr doc for record: id=6bdc3c1d-d21a-43e3-aa84-baeeda601bb3

ERROR: [ConcurrentUpdateSolrClient] - errororg.apache.solr.client.solrj.impl.HttpSolrClient$RemoteSolrException: 
Error from server at http://localhost:8983/solr/biocache: Conflictrequest: http://localhost:8983/solr/biocache/update?commit=true&softCommit=false&waitSearcher=false&_version_=1&failOnVersionConflicts=false&wt=javabin&version=2›

Remote error message: Document not found for update.  id=c0ee1a86-1df6-40b2-950c-bdde40b1c46e_invalid at org.apache.solr.client.solrj.impl.ConcurrentUpdateSolrClient$Runner.sendUpdateStream  

 

The error is Document not found for update which is as expected. Out of the 4 updates the 2nd one has an invlid id so what I expected is 2nd update failed but all the other 3 are updated successfully (since I set failOnVersionConflicts: false).

But the reality is, only 1st update succeeded.  All others failed because an exception was thrown for 2nd update.

I then went on to check the 8.5.1 sourcecode, the exception is thrown from getUpdatedDocument 

 

SolrInputDocument oldRootDocWithChildren = RealTimeGetComponent.getInputDocument(cmd.getReq().getCore(), idBytes, RealTimeGetComponent.Resolution.ROOT_WITH_CHILDREN);

if (oldRootDocWithChildren == null) {
 if (versionOnUpdate > 0) {
 // could just let the optimistic locking throw the error
 throw new SolrException(ErrorCode.CONFLICT, "Document not found for update. id=" + idString);
 } else if (req.getParams().get(ShardParams._ROUTE_) != null) {

It tries to find a doc with the specified id (null is found in this case), then seeing versionOnUpdate == 1, an exception is thrown anyway. But where is the failOnVersionConflicts **used ?

 

it's in doVersionAdd(). In this function we call getUpdateDocument first in line374, then we compare versionOnUpdate with foundVersion and the failOnVersionConflitcs flag. In my case there's no chance to check the flag because an exception is directly thrown before that.

 

getUpdatedDocument(cmd, versionOnUpdate);

// leaders can also be in buffering state during "migrate" API call, see SOLR-5308
if (forwardedFromCollection && ulog.getState() != UpdateLog.State.ACTIVE
 && isReplayOrPeersync == false) {
 // ....
 return true;
}

if (versionOnUpdate != 0) {
 Long lastVersion = vinfo.lookupVersion(cmd.getIndexedId());
 long foundVersion = lastVersion == null ? -1 : lastVersion;
 if (versionOnUpdate == foundVersion || (versionOnUpdate < 0 && foundVersion < 0)
 || (versionOnUpdate == 1 && foundVersion > 0)) {
 // we're ok if versions match, or if both are negative (all missing docs are equal), or if cmd
 // specified it must exist (versionOnUpdate==1) and it does.
 } else {
 if(cmd.getReq().getParams().getBool(CommonParams.FAIL_ON_VERSION_CONFLICTS, true) == false) {
 return true;
 }

It looks like an issue. 

 
 =======  update ==========

according to solr documentation 

[Updating Parts of Documents | Apache Solr Reference Guide 8.5|https://solr.apache.org/guide/8_5/updating-parts-of-documents.html]
 

$ curl -X POST -H 'Content-Type: application/json' 'http://localhost:8983/solr/techproducts/update?versions=true&_version_=-1&failOnVersionConflicts=false&omitHeader=true' --data-binary ' [ { "id" : "aaa" }, { "id" : "ccc" } ]'

In this example, we have added 2 documents "aaa" and "ccc". As we have specified the parameter _version_=-1, this request should not add the document with the id aaa because it already exists. The request succeeds & does not throw any error because the failOnVersionConflicts=false parameter is specified

If solr can recognize _version= -1 and failOnVersionConflicts=false and then only updates those not existing. There's no reason it can't do similar to version_ = 1

Issue Links

• links to: GitHub Pull Request #150