Thread

Re: BUG #19348: Disk space error when running COPY after upgrading to PostgreSQL 17.7

Thomas Munro <thomas.munro@gmail.com> — 2025-12-09T21:19:47Z

On Wed, Dec 10, 2025 at 12:32 AM PG Bug reporting form
<noreply@postgresql.org> wrote:
> The following bug has been logged on the website:
>
> Bug reference:      19348
> Logged by:          weijie JL
> Email address:      weijie1006jl@gmail.com
> PostgreSQL version: 17.7
> Operating system:   RockyLinux8
> Description:
>
> After upgrading to PostgreSQL 17.7, our scheduled data synchronization task
> intermittently fails. About 6 seconds into the run, during the COPY import
> stage, we get the following error:
>
> ERROR: 53100: could not extend file "base/16416/4201185432" with
> FileFallocate(): No space left on device
> However, the disk has ample free space. We see no abnormal data size growth,
> and the table involved is small — roughly a dozen columns, each containing
> short text strings of only a few dozen characters.

Could this be a case of the mysterious ENOSPC reported by XFS since we
started using posix_fallocate() in COPY?  IIRC that's Rocky's default
file system.  I'm working on a patch that would allow the next minor
releases to turn that off, originally for the benefit of BTRFS
compression users but it was pointed out[1] that XFS users might want
this too until the root cause can be figured out.  (Passing thought: I
don't recall anyone reproducing this on Debian systems, I know I
tried... could RHEL/Rocky have a local difference?)

[1] https://www.postgresql.org/message-id/flat/CAKZiRmzF6SRR%2BAxPZsXzs1kaHrxjLO2sbLT4Q61EMA5PGQB3Vg%40mail.gmail.com#b07d8e3bcef37e9afda96e0333cb1454