Dear all,
We are running Xrootd 5.4.2-3
in an redirector
(xrootd.phy.bris.ac.uk:1094) → gateway
(io-37-02.acrc.bris.ac.uk:1194) setup on CentOS 7 and connect storage via xrootd-hdfs
(async disabled).
We offer xrootd
and https-over-xrootd
, our issue is mostly for the latter.
The issue we are experiencing can be summarized in the mermaid graph:
sequenceDiagram
participant Client
participant XrootD Redirector
participant XrootD Gateway
participant HDFS
Client ->> XrootD Redirector: rename /path/to/file /path/to/file.new
XrootD Redirector ->> XrootD Gateway: rename /path/to/file /path/to/file.new
XrootD Gateway ->> HDFS: rename /path/to/file /path/to/file.new
HDFS -->> XrootD Gateway: rename done
Note left of Client: Waits 5 seconds
Client ->> XrootD Redirector: stat /path/to/file.new
XrootD Redirector ->> Client: /path/to/file.new does not exist
and is explained in more detail in #235 (comment).
In short: file is renamed on disk, but redirector thinks it is not there.
This is not an issue for most VOs, but operational tests (OPS VO) are failing - very bad.
Since this is unlikely to be solved internally in xrootd (based on the comment in issue 235), I was wondering if it would be possible to solve it at config level.
Can we change stat/propfind requests to
always query the gateway and file system:
sequenceDiagram
participant Client
participant XrootD Redirector
participant XrootD Gateway
participant HDFS
Client ->> XrootD Redirector: stat /path/to/file.new
XrootD Redirector ->> XrootD Gateway: stat /path/to/file.new
XrootD Gateway ->> HDFS: stat /path/to/file.new
HDFS -->> XrootD Gateway: stat results
XrootD Gateway -->> Client: stat results
or can we change the redirector to directly query the file system (read-only):
sequenceDiagram
participant Client
participant XrootD Redirector
participant HDFS
Client ->> XrootD Redirector: stat /path/to/file.new
XrootD Redirector ->> HDFS: stat /path/to/file.new
HDFS -->> XrootD Redirector: stat results
XrootD Redirector -->> Client: stat results
The particular test we are failing: argo-mon - not sure if this is available to everyone.
Apparently, more info on that test is on a twiki
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1