Uploaded image for project: 'Confluence Data Center'
  1. Confluence Data Center
  2. CONFSERVER-34383

Severe Confluence Server Crash Down On Word Document Import

    • Icon: Bug Bug
    • Resolution: Fixed
    • Icon: High High
    • 5.9.1
    • 5.5.2, 5.6.4

      While importing a particular word document into Confluence we faced repetitively a severe confluence server crash down, meaning that it is not operational and reachable anymore! We managed to isolate the content in the word document that produces the severe malfunction to be a particular image. Copying this image from a .docx file to a .doc file will produce the same error. However, modifying the bit pattern of this image slighty (e.g., by switching to another compression level) will solve the problem. Of course, this is a very unpleasant situation because if anybody tries to import another word document (with comparable "features") our server will crash down and the current work of many other users might be lost. Because we are seriously considering to upgrade our starter license to a 500 user license, this issue needs definitely to be solved in advance.

      In parallel, due to the priority level we have already submitted a support request under the same summary.

      Thank you very much in advance for a fast solution!

      Steps to Reproduce

      1. In Confluence 5.5.2, create a page and go to Tools > Import Word Doc
      2. Import Testfall.docx, leaving all options as default
      3. Observe that the Confluence process is terminated.
      4. This is all that appears in the logs:
        2014-08-18 12:51:24,510 WARN [http-bio-10552-exec-7] [atlassian.confluence.cache.TransactionalCacheFactory] warning Transactional cache update outside transaction. All updates to this cache should be performed from a thread with a valid transaction context.
         -- referer: http://localhost:10552/confluence/pages/worddav/uploadimport.action?pageId=950274 | url: /confluence/rest/stp/1.0/license/status | userName: admin
        2014-08-18 12:51:30,781 WARN [http-bio-10552-exec-9] [com.benryan.webwork.WordImportAction] executeFileUpload Failed to delete uploaded file /home/denise/confluence/home/552home/temp/upload_bdfcea9b_ccf0_4f35_a28f_2e94ad861811_00000001.tmp
         -- url: /confluence/pages/worddav/importword.action | page: 950274 | userName: admin | referer: http://localhost:10552/confluence/pages/worddav/uploadimport.action?pageId=950274 | action: importword
        2014-08-18 12:51:31,276 WARN [http-bio-10552-exec-5] [atlassian.confluence.cache.TransactionalCacheFactory] warning Transactional cache update outside transaction. All updates to this cache should be performed from a thread with a valid transaction context.
         -- referer: http://localhost:10552/confluence/pages/worddav/importword.action | url: /confluence/rest/stp/1.0/license/status | userName: admin
        
      5. Running strace on the process shows that it is killed by a seg fault:
        Process 4901 attached
        futex(0x7fe36b4269d0, FUTEX_WAIT, 4905, NULL <unfinished ...>
        +++ killed by SIGSEGV (core dumped) +++
        

      Workaround

      Save the image in another format and use that instead. I used .bmp for the attached testfall-working.doc, which imports successfully.

      If you regularly experience this problem, and are running Confluence 5.9.1 or later, see JVM crashes during Import from Word knowledge base article for details of a workaround.

        1. Confluence_support_2014-07-25-08-47-48.zip
          3.89 MB
        2. Testfall.doc
          66 kB
        3. Testfall.docx
          57 kB
        4. testfall-working.doc
          179 kB

            [CONFSERVER-34383] Severe Confluence Server Crash Down On Word Document Import

            Reopening because we are looking at the feasibility of a workaround.

            Denise Unterwurzacher [Atlassian] (Inactive) added a comment - Reopening because we are looking at the feasibility of a workaround.

            I totally agree with @Nicholas Randal. During the past years, we often experienced issues with Word Import, Office Connector and PDF/HTML/XML exports. These actions crashed our system often and we never received a sufficient solution by Atlassian. Therefore, we always had to disable the features via Core Change or build size limits inside (again via Core Change).

            We would love to see Atlassian working on reliable & stable Office Connector and Export functions.

            Jens Kasperek (Bosch GmbH) (Inactive) added a comment - I totally agree with @Nicholas Randal. During the past years, we often experienced issues with Word Import, Office Connector and PDF/HTML/XML exports. These actions crashed our system often and we never received a sufficient solution by Atlassian. Therefore, we always had to disable the features via Core Change or build size limits inside (again via Core Change). We would love to see Atlassian working on reliable & stable Office Connector and Export functions.

            Nicholas Randal added a comment - - edited

            I did that already. JST-143746
            Importing word should probably not cause confluence to become completely unresponsive. I understand that a failure can happen on import and an error message was shown. However the process ran away at that point slamming the CPU to 100%. We had to shutdown the server through remote command, as we could no longer remote into the machine. Confluence was totally consuming all resources on the system, not leaving enough to actually remote in to the system. This is a show stopping problem. I was in the middle of a demo to all our major stakeholders showing why we should move away from product X to Confluence. Imagine my embarrassment when the whole system went down and would not recover.

            Nicholas Randal added a comment - - edited I did that already. JST-143746 Importing word should probably not cause confluence to become completely unresponsive. I understand that a failure can happen on import and an error message was shown. However the process ran away at that point slamming the CPU to 100%. We had to shutdown the server through remote command, as we could no longer remote into the machine. Confluence was totally consuming all resources on the system, not leaving enough to actually remote in to the system. This is a show stopping problem. I was in the middle of a demo to all our major stakeholders showing why we should move away from product X to Confluence. Imagine my embarrassment when the whole system went down and would not recover.

            nrandal Please open a support request at https://support.atlassian.com so we can help get your demo working again.

            Ann Worley (Inactive) added a comment - nrandal Please open a support request at https://support.atlassian.com so we can help get your demo working again.

            Experiencing the same issue. However, I cannot get the server to stabilize. I am in the middle of trying to demo Confluence to my company, so this is hitting at the worst possible time. During a demo I tried to import a word document. This failed and is slamming the CPU to 100% and causing the whole server to become unstable.

            At this point I don't care if the document imports but I don't have a working demo anymore to show.

            Nicholas Randal added a comment - Experiencing the same issue. However, I cannot get the server to stabilize. I am in the middle of trying to demo Confluence to my company, so this is hitting at the worst possible time. During a demo I tried to import a word document. This failed and is slamming the CPU to 100% and causing the whole server to become unstable. At this point I don't care if the document imports but I don't have a working demo anymore to show.

            Today, we 'experienced' this bug. A user tried to upload a word file with an embedded visio object. This led to error messages on log:
            _2015-06-08 13:54:18,189 WARN [http-bio-8080-exec-1586] [com.benryan.webwork.WordImportAction] executeFileUpload Failed to delete uploaded file /projects/confluence/home-p-5.5.6/temp/upload_6ec28a59_de1c_4c73_a98a_193bae650669_00045199.tmp
            – url: /confluence/pages/worddav/importword.action | page: 338664955 | userName: abcdef | referer: https://***.***.com/confluence/pages/worddav/uploadimport.action?pageId=338664955 | action: importword_

            On JAVA Melody monitoring we saw endless running requests:
            /pages/worddav/processwarn.action POST

            We are running Confluence 5.5.6 with these system parameters:
            Operating System Linux 2.6.32-504.8.1.el6.x86_64
            OS Architecture amd64
            Application Server Apache Tomcat/7.0.47
            Servlet Version 3.0
            Java Version 1.7.0_75
            Java Vendor Oracle Corporation
            JVM Version 1.7
            JVM Vendor Oracle Corporation
            JVM Implementation Version 24.75-b04
            Java Runtime Java(TM) SE Runtime Environment
            Java VM Java HotSpot(TM) 64-Bit Server VM

            I do not know why this issue has been marked as resolved, if there is no solution mentioned? And how should we pretend users to upload word images with images or office objects?

            Jens Kasperek (Bosch GmbH) (Inactive) added a comment - - edited Today, we 'experienced' this bug. A user tried to upload a word file with an embedded visio object. This led to error messages on log: _2015-06-08 13:54:18,189 WARN [http-bio-8080-exec-1586] [com.benryan.webwork.WordImportAction] executeFileUpload Failed to delete uploaded file /projects/confluence/home-p-5.5.6/temp/upload_6ec28a59_de1c_4c73_a98a_193bae650669_00045199.tmp – url: /confluence/pages/worddav/importword.action | page: 338664955 | userName: abcdef | referer: https://***.***.com/confluence/pages/worddav/uploadimport.action?pageId=338664955 | action: importword_ On JAVA Melody monitoring we saw endless running requests: /pages/worddav/processwarn.action POST We are running Confluence 5.5.6 with these system parameters: Operating System Linux 2.6.32-504.8.1.el6.x86_64 OS Architecture amd64 Application Server Apache Tomcat/7.0.47 Servlet Version 3.0 Java Version 1.7.0_75 Java Vendor Oracle Corporation JVM Version 1.7 JVM Vendor Oracle Corporation JVM Implementation Version 24.75-b04 Java Runtime Java(TM) SE Runtime Environment Java VM Java HotSpot(TM) 64-Bit Server VM I do not know why this issue has been marked as resolved, if there is no solution mentioned? And how should we pretend users to upload word images with images or office objects?

            Minh Tran added a comment -

            Dear all,

            We carefully analysed this ticket within our team and figured out this problem happened because of a bug in JDK (Java Development Kit)
            The bug is being tracked here: https://bugs.openjdk.java.net/browse/JDK-6996336
            When this issue is resolved you can upgrade to the latest JDK to apply the fix

            Please let me know if you still have any concerns

            Thanks & Regards,
            Minh Tran
            Confluence BugMaster
            Atlassian

            Minh Tran added a comment - Dear all, We carefully analysed this ticket within our team and figured out this problem happened because of a bug in JDK (Java Development Kit) The bug is being tracked here: https://bugs.openjdk.java.net/browse/JDK-6996336 When this issue is resolved you can upgrade to the latest JDK to apply the fix Please let me know if you still have any concerns Thanks & Regards, Minh Tran Confluence BugMaster Atlassian

            Kenny MacLeod added a comment - - edited

            I can confirm that this still occurs with the latest Confluence 5.7 SNAPSHOT on jdk 1.8.0_25 under OSX. It crashes with "EXC_CRASH (SIGILL)". The problem seems specific to the "Word Import" function - uploading the same file as an attachment and allowing it to index does not cause any issues.

            The most suspicious part of the OSX stack dump that I get back comes from an HTTP thread which looks like it's stuck in a recursive loop in something called processCubic in libdcpr.dynlib. That looks like some kind of graphic manipulation to me, possibly due to trying to process images in the doc file.

            Possibly related: JVM crash in recursive processCubic calls (libdcpr.so) - it that's the same thing, then it's a Java2D bug in the JVM that Oracle have yet to fix.

            Kenny MacLeod added a comment - - edited I can confirm that this still occurs with the latest Confluence 5.7 SNAPSHOT on jdk 1.8.0_25 under OSX. It crashes with "EXC_CRASH (SIGILL)". The problem seems specific to the "Word Import" function - uploading the same file as an attachment and allowing it to index does not cause any issues. The most suspicious part of the OSX stack dump that I get back comes from an HTTP thread which looks like it's stuck in a recursive loop in something called processCubic in libdcpr.dynlib . That looks like some kind of graphic manipulation to me, possibly due to trying to process images in the doc file. Possibly related: JVM crash in recursive processCubic calls (libdcpr.so) - it that's the same thing, then it's a Java2D bug in the JVM that Oracle have yet to fix.

            Hi sven-ole.voigt

            Thanks for getting back in touch. This is currently on the backlog awaiting a developer to address it. As soon as we have any more information on this we'll update this ticket with details.

            Regards
            Steve Haffenden
            Confluence Bugmaster
            Atlassian

            Steve Haffenden (Inactive) added a comment - Hi sven-ole.voigt Thanks for getting back in touch. This is currently on the backlog awaiting a developer to address it. As soon as we have any more information on this we'll update this ticket with details. Regards Steve Haffenden Confluence Bugmaster Atlassian

            Thanks for your reply. However, despite the higher priority level, currently, there seems to be no further progress. Could you tell me how the current status is?

            Sven-Ole Voigt added a comment - Thanks for your reply. However, despite the higher priority level, currently, there seems to be no further progress. Could you tell me how the current status is?

              ple Phong Quoc Le (Inactive)
              c4a16aca0136 Sven-Ole Voigt
              Affected customers:
              1 This affects my team
              Watchers:
              13 Start watching this issue

                Created:
                Updated:
                Resolved: