FS#134 - Package Installs from trunk cause random incomplete installs/failures

[aparcar]

jon.grossart:

I tried doing a sysupgrade to r1480 (2016-09-01) yesterday on a C2600, and it proved very difficult to get back to a full install. The actual image worked correctly, but there were random failures when installing packages from the feeds. r1382 (2016-8-19) was the last build I had used and packages had worked correctly (minus some odd wifi stalling bug). This was confirmed by another user as well. The failures happened regardless of installing them all together or installing them in chunks.

Packages installed post boot:luci, luci-theme-material, luci-app-uhttpd, luci-app-upnp, ddns-scripts, ddns-scripts_no-ip_com, luci-app-ddns, sqm-scripts, luci-app-sqm, openssh-sftp-server, nano

I tried installing various times (about 5) across different base systems (r1261, r1382 although of course sqm would fail because of kernel versions), and there would be different failure symptoms each time including:

• luci won't start with "Bad Gateway" (/www/cgi-bin/luci would show various errors)

• luci theme shows up as unformatted plain HTML rather than with theme -- sometimes this would affect bootstrap, sometimes material

• luci would say libiwinfo-lua wasn't installed when it was, so it wouldn't show wireless info

• Configuring ddns-scripts. /etc/rc.common: .: line 143: can't open '/usr/lib/ddns/dynamic_dns_functions.sh' Configuring ddns-scripts_no-ip_com. //usr/lib/opkg/info/ddns-scripts_no-ip_com.postinst: /usr/lib/opkg/info/ddns-scripts_no-ip_com.postinst-pkg: line 2: can't create /usr/lib/ddns/services: nonexistent directory

• would sometimes get collected errors about files not existing for various package Collected errors:

  • pkg_get_installed_files: Failed to open //usr/lib/opkg/info/luci-proto-ppp.list: No such file or directory.

• SSH prompt would change from "root@lede:~#" to "root@lede:(unreachable)/root#" after the presumed failure

Force reinstalling the packages (if I could determine the messed up ones) would fix the issues. Since the downloads are completing fine, it would seem that there is some sort of error actually installing the packages/writing to flash. However, even when there are errors, it marks the packages as installed correctly.

The other user who checked this bug had no problems with luci when compiled into his personal build.

[aparcar]

jon.grossart:

For reference, I tried twice on r1491 x86-64 image in Virtualbox, and the packages all installed correctly and worked as expected. So it's not a global problem.

[aparcar]

dissent1:

I confirm the same issue on Netgear r7800 using trunk images - each time random package fails to install. Precompiled works fine though.

[aparcar]

jon.grossart:

Tried on the 2016-09-09 snapshot and it appears to be working again. I'm not sure if an intermediate bug was fixed or if there was some network infrastructure problem, but it seems to working now. Needs a confirm from dissent1 on his device and I'll close this bug.

[aparcar]

jow-:

That was likely related to switching opkg to vfork which triggered multiple broken code paths. We reverted it to its old behaviour for now.