That request had already rolled off, so I had to search for a new one. Here's this:$ kubectl --namespace=devpi-server logs devpi-server-primary-85d6484cb7-hgqpp | grep req86032020-01-03 13:22:19,560 INFO: [req8603] GET /root/pypi/+simple/internal_project_name/2020-01-03 13:22:19,631 ERROR: [req8603] [Rtx14627] while handling https://primary-devpi-server.example.com/root/pypi/+simple/internal_project_name/:I'm trying to troubleshoot that further, so I grepped for the URI, which nothing should be requesting:$ kubectl --namespace=devpi-server logs devpi-server-primary-85d6484cb7-hgqpp | grep "/root/pypi/+simple/internal_project_name/"2020-01-02 18:10:53,164 INFO: [req82] GET /root/pypi/+simple/internal_project_name/2020-01-02 18:10:53,170 ERROR: [req82] [Rtx14604] while handling https://primary-devpi-server.example.com/root/pypi/+simple/internal_project_name/:"172.27.23.69" (172.27.23.69) - [02/Jan/2020:18:10:53 +0000] "GET /root/pypi/+simple/internal_project_name/ HTTP/1.1" 404 1037 "-" 0.013 "http/https/https" "primary-devpi-server.example.com/primary-devpi-server.example.com/primary-devpi-server.example.com" "80/443/443"That IP address from which that request came is the IP address of one of the mirror servers. When I grep the mirror logs for that internal project name, there is nothing indicating that it is looking in root/pypi; only org_name/stable:$ kubectl --namespace=devpi-server logs devpi-server-replica-0 | grep internal_project_name2020-01-02 18:10:53,150 INFO: [req37] GET /org_name/stable/+simple/internal_project_name/"172.27.36.103" (172.27.36.103) - [02/Jan/2020:18:10:53 +0000] "GET /org_name/stable/+simple/internal_project_name/ HTTP/1.1" 200 1878 "-" 0.101 "http/https/https" "devpi.example.com/devpi.example.com/devpi.example.com" "80/443/443"So our builds (locally and on Jenkins) are requesting /org_name/stable/+simple/internal_project_name/ of the mirrors, but then the mirrors are requesting /root/pypi/+simple/internal_project_name/ of the master. But they must at some point be requesting the correct URL, because the mirrors have all of the proper projects in org_name/stable, and you can install them all just fine using the mirrors.On the master, the indexes are configured as follows (org_name/stable extends org_name/upstream extends root/pypi):$ devpi index -v org_name/stabletype=stagebases=org_name/upstreamvolatile=Falseacl_upload=org_nameacl_toxresult_upload=org_namemirror_whitelist=$ devpi index -v org_name/upstreamtype=stagebases=root/pypivolatile=Falseacl_upload=org_nameacl_toxresult_upload=org_namemirror_whitelist=*The mirror/replica servers are configured like this:devpi-server:serverdir: /data/devpi-server/server-datasecretfile: /data/devpi-server/server-data/.secretunix-socket: /tmp/devpi-server.sockunix-socket-perms: "0666"threads: 16replica-max-retries: 2request-timeout: 5root-passwd: ******************theme: semantic-uirole: replicamaster-url: https://primary-devpi-server.example.comHope that additional information helps.Thanks,NickOn Fri, Jan 3, 2020 at 3:32 AM Florian Schulze <mail@florian-schulze.net> wrote:On 2 Jan 2020, at 18:19, Nicholas Williams wrote:
> We're starting to roll out our new Devpi installation to projects now
> that
> we've tested it extensively. We have one master and three mirrors. For
> the
> purposes of this email, the mirrors are load-balanced behind the
> domain
> name "devpi.example.com," and the master is available at "
> primary-devpi-server.example.com."
>
> We configured our Pipenv and Pip installations to use this index URL:
> https://devpi.example.com/org_name/stable/+simple/
>
> That URL will use the mirrors. The "org_name/stable" index extends
> root/pypi and also includes our internal projects. Everything appears
> to be
> working on the client. No errors, installs are going off without a
> hitch.
>
> However, the Devpi server on the master machine (not the mirrors) logs
> a
> ton of errors like the following:
>
> [req78237] [Rtx14602] while handling
> https://primary-devpi-server.example.com/root/pypi/+simple/internal_project_name/
> :
> The project internal_project_name does not exist.
>
> What is the cause of these errors? Is this something that should
> concern
> us, or something that we can ignore? "internal_project_name" installs
> just
> fine, but for some reason yields those errors on the master server.
Is there more log output with the same request number ("req78237" in
this case)? Without that it is hard to say. Looking at the code this
should only happen if root/pypi is accessed directly. What is the
setting of ``mirror_whitelist`` in "org_name/stable"? I tried with an
empty whitelist and with * and in both cases I wasn't able to provoke
the error for a private package.
Regards,
Florian Schulze