[
https://issues.apache.org/jira/browse/CAMEL-25859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18131398#comment-18131398
]
Claus Ibsen commented on CAMEL-25859:
-------------------------------------
Merged to main in https://github.com/apache/camel/pull/28381 (commit
9141da28a7da), fixed in 4.24.0.
_Claude Code on behalf of davsclaus_
> camel-mongodb-gridfs - the consumer routes a file with the content and
> headers of the newest file that has the same name, so the content of the
> older file is never routed
> --------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-25859
> URL: https://issues.apache.org/jira/browse/CAMEL-25859
> Project: Camel
> Issue Type: Bug
> Components: camel-mongodb-gridfs
> Reporter: shashank
> Assignee: shashank
> Priority: Major
> Fix For: 4.24.0
>
>
> {{GridFsConsumer.processCollection}} (main 20cdb309f7d0, line 212) opens the
> download stream of each file the poll found by its file name:
> {code:java}
> GridFSDownloadStream downloadStream =
> endpoint.getGridFsBucket().openDownloadStream(file.getFilename());
> file = downloadStream.getGridFSFile();
> {code}
> GridFS does not make file names unique: every upload with a name that already
> exists adds a file (a revision), and {{GridFSBucket.openDownloadStream(String
> filename)}} opens the newest revision of that name. The component docs say so
> too (the {{count}} operation counts the files matching a filename,
> {{listAll}} lists names with their ids). So when one poll finds two or more
> files with the same name, every exchange gets the stream of the newest one,
> and the headers ({{CamelFileLength}}, {{CamelFileLastModified}}, content
> type, metadata) are taken from it as well. The older file is routed with the
> content of the newer one, its own content is never routed, and the newer
> content is routed twice. With a file attribute strategy the older file is
> then marked {{done}}.
> When does a poll find two files with the same name: two uploads of the same
> name within one poll delay (500 ms by default); with {{PersistentTimestamp}},
> every revision uploaded while the consumer was stopped; with the
> {{FileAttribute}} strategies, every revision not marked yet, for example all
> of them at the first start. A producer that uploads a report or an export
> under a fixed name ({{CamelFileName}} header) does exactly this.
> *Fix*: open the stream by the id of the found file
> ({{openDownloadStream(file.getId())}}). The query, the file attribute and the
> timestamp already work by id and upload date; only the stream was by name.
> *Tests*: {{GridFsConsumerTimestampTest}} (the fake bucket of CAMEL-25607 /
> CAMEL-25711) now answers {{openDownloadStream}} as MongoDB does: by name the
> newest file of that name, by id that file. New
> {{filesWithTheSameNameAreEachRoutedWithTheirOwnContent}}, for each of the 5
> query strategies: two files named {{report.csv}} found by the same poll. On
> main all 5 fail: {{expected: <[<id of the first>, <id of the second>]> but
> was: <[<id of the second>, <id of the second>]>}}. With the fix they pass;
> camel-mongodb-gridfs unit tests: 20 run, 0 failures (the ITs need Docker and
> were not run).
> *How it was found*: a review of the consumers that run their own poll loop;
> the stream opened by name next to a query and a file attribute that work by
> id stood out. Then the JUnit test above. Not run against a MongoDB server;
> the by-name behaviour is that of the documented driver API.
> *Affected*: main, camel-4.23.x, camel-4.22.x, camel-4.18.x and camel-4.14.x
> (the same {{openDownloadStream(file.getFilename())}} in {{GridFsConsumer}};
> read, not run).
> _Filed with Claude Code on behalf of allthingssecurity._
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
