Production postmortemThe file system limitation

time to read 2 min | 221 words

On an otherwise uneventful morning, the life of the operations guy got… interesting.

What were supposed to be a routine morning got hectic because the database refused to operate normally. To be more exact, the database refused to load a file. RavenDB is generally polite when it run into issues, but this time, it wasn’t playing around. Here is the error it served:

---> System.IO.IOException: Could not set the size of file  D:\RavenData\Databases\Purple\Raven.voron to 820 GBytes

---> System.ComponentModel.Win32Exception (665): The requested operation could not be completed due to a file system limitation

Good old ERROR_FILE_SYSTEM_LIMITATION, I never knew you, because we have never run into an error with this in the past.

The underlying reason was simple, we had a large file (820GB) that was too fragmented. At some point, the number of fragments of the file bypassed the maximum size of the file system.

The KB article about this issue is here. You might be able to move forward more quickly by using the contig.exe tool to defrag a single file.

The root cause here was probably backing up to the same drive as the database, which forced the file system to break the database file into fragements.

Just a reminder that there are always more layers into the system and that we need to understand them all when they break.

More posts in "Production postmortem" series:

  1. (17 Sep 2021) The Guinness record for page faults & high CPU
  2. (07 Jan 2021) The file system limitation
  3. (23 Mar 2020) high CPU when there is little work to be done
  4. (21 Feb 2020) The self signed certificate that couldn’t
  5. (31 Jan 2020) The slow slowdown of large systems
  6. (07 Jun 2019) Printer out of paper and the RavenDB hang
  7. (18 Feb 2019) This data corruption bug requires 3 simultaneous race conditions
  8. (25 Dec 2018) Handled errors and the curse of recursive error handling
  9. (23 Nov 2018) The ARM is killing me
  10. (22 Feb 2018) The unavailable Linux server
  11. (06 Dec 2017) data corruption, a view from INSIDE the sausage
  12. (01 Dec 2017) The random high CPU
  13. (07 Aug 2017) 30% boost with a single line change
  14. (04 Aug 2017) The case of 99.99% percentile
  15. (02 Aug 2017) The lightly loaded trashing server
  16. (23 Aug 2016) The insidious cost of managed memory
  17. (05 Feb 2016) A null reference in our abstraction
  18. (27 Jan 2016) The Razor Suicide
  19. (13 Nov 2015) The case of the “it is slow on that machine (only)”
  20. (21 Oct 2015) The case of the slow index rebuild
  21. (22 Sep 2015) The case of the Unicode Poo
  22. (03 Sep 2015) The industry at large
  23. (01 Sep 2015) The case of the lying configuration file
  24. (31 Aug 2015) The case of the memory eater and high load
  25. (14 Aug 2015) The case of the man in the middle
  26. (05 Aug 2015) Reading the errors
  27. (29 Jul 2015) The evil licensing code
  28. (23 Jul 2015) The case of the native memory leak
  29. (16 Jul 2015) The case of the intransigent new database
  30. (13 Jul 2015) The case of the hung over server
  31. (09 Jul 2015) The case of the infected cluster