I can confirm this issue happening also in my local development
environment. I had ~30 network interfaces (don't ask why) which caused
fairly long filenames generated for persistent store path. I figured it
out while debugging startup issue of tests. Arthur pointed out his
earlier mail, but I am not entirely sure if IGNITE-16938 is related.
Thinking of consistent node identifier and its storage path, are these
IP addresses in generated path are relevant for third party tooling? If
not, why not getting host addresses, sorting them and then calculating a
hash? It would be consistent, short and meaningless.
Best,
Łukasz
On 27.10.2022 19:12, Arthur Naseef wrote:
Forwarding this one. Looks like there is a new problem introduced to
the code base since 2.14 here.
We are running into it again while working on the current development
branch.
Art
---------- Forwarded message ---------
From: *Arthur Naseef* <artnas...@apache.org <mailto:artnas...@apache.org>>
Date: Wed, Jul 6, 2022 at 5:14 PM
Subject: ENAMETOOLONG on mkdir() leading to failed startup with
exception org.apache.ignite.IgniteCheckedException: Failed to create
page store work directory: ...
To: <u...@ignite.apache.org <mailto:u...@ignite.apache.org>>
I started using the latest master brannch in local testing of Apache
Ignite and ran into the following exception (note IP addresses and path
elements replaced redacted with asterisks to prevent possible security
concerns):
Caused by: org.apache.ignite.IgniteCheckedException: Failed to
create page store work directory:
/home/art/****/******/*******/*******/POC/ignite/distributed-scheduling-poc/ignite/work/db/*_*_*_*_*_*_*_*_**_***_*_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_***_*_***_***_***_***_*_***_***_**_*_***_***_***_*_***_***_***_*_***_***_***_*_***_***_***_*_***_***_**_*_***_***_**_*_****_****_***_****_*_*_*_*_**_************_*****
Using strace, found the following underlying cause:
mkdir("/home/art/****/******/*******/*******/POC/ignite/distributed-scheduling-poc/ignite/work/db/*_*_*_*_*_*_*_*_**_***_*_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_**_*_*_***_***_*_***_***_***_***_*_***_***_**_*_***_***_***_*_***_***_***_*_***_***_***_*_***_***_***_*_***_***_**_*_***_***_**_*_****_****_***_****_*_*_*_*_**_************_*****",
0777) = -1 ENAMETOOLONG (File name too long)
Debugging through the code, I *THINK* the following change may be involved:
ticket https://issues.apache.org/jira/browse/IGNITE-16938
<https://issues.apache.org/jira/browse/IGNITE-16938>
commit
https://github.com/apache/ignite/commit/cdf8200fa02da8e88b6ca5a9fd9e89bb28f95c62
<https://github.com/apache/ignite/commit/cdf8200fa02da8e88b6ca5a9fd9e89bb28f95c62>
It appears the path name includes all of the IP address of all the
interfaces on the system, and since this system is running docker and
kubernetes, the number of those interfaces is ... large.
I'll go back to a released version for my DEV purposes for now.
Feedback is welcome.
Art
