Ok, the failures were the same with fcgi and gunicorn app servers. But it _works_ with django's dev server (which you must not use in production due to memory/security holes etc.). While ConTeXt runs it looks like: 5852 25767 TS 21 18:04 pts/0 00:00:00 \_ python manage.py runserver 8000 1477 5852 TS 23 18:07 pts/0 00:00:03 \_ /var/www/ cerebrale/bin/python manage.py runserver 8000 3395 1477 TS 22 18:08 pts/0 00:00:00 \_ /bin/sh -c date "+%Y-%m-%d %H:%M:%S" && cd /var/www/.../_textemp && /var/opt/ context/tex/texmf-linux-64/bin/context --purgeall --batchmode -- result="/var/www/... 3397 3395 TS 22 18:08 pts/0 00:00:00 \_ /bin/ sh /var/opt/context/tex/texmf-linux-64/bin/context --purgeall -- batchmode --result=/var/www/... 3398 3397 TS 15 18:08 pts/0 00:00:00 \_ texlua /var/opt/context/tex/texmf-linux-64/bin/mtxrun --script context --purgeall --batchmode --result=/var/www/... 3400 3398 TS 21 18:08 pts/0 00:00:00 \_ luatex --interaction=batchmode --fmt=/var/opt/context/tex/texmf- cache/.../cont-en --lua=/var/opt/context/tex... TeX completes fast, and everything seems well afterwards. But I started that app server from an active shell. Maybe that's the difference? Right, if I start gunicorn manually, it also works. So, the culprit seems to be the automatic background start of the app server via daemontools. I'll have to test other means of starting and supervising the processes - e.g. runit or supervisor. Since I like Python and don't want to replace "init" yet, I try the latter first. Some hours later: YES IT WORKS with supervisor so far! Thank you all for your patience, ideas and help! Greetlings from Lake Constance! Hraban --- http://www.fiee.net/texnique/ http://wiki.contextgarden.net https://www.cacert.org (I'm an assurer)