Setting pipefail for a single piped command

From the Bash command line you would need to invoke a subshell to avoid pipefail being set afterwards:

$ (set -o pipefail && command1 | command2 | command3)

This would limit the effect of the pipefail option to the subshell create by the parentheses (...).

A real example:

$ (set -o pipefail && false | true) && echo pipefail inactive || echo pipefail active
pipefail active

If you use an explicit shell call with the -c option you do not need a subshell, either with bash or with an sh alias to bash:

$ bash -c "set -o pipefail && false | true" && echo pipefail inactive || echo pipefail active
pipefail active
$ sh -c "set -o pipefail && false | true" && echo pipefail inactive || echo pipefail active
pipefail active

Since your sh does not accept the pipefail option, I would have to assume that it is either some older or modified version of bash - or that it is actually some other shell completely.

Not sure why it wasn't mentioned above, but it is possible to explicitly unset pipefail, using set +o pipefail.

set -o pipefail
command1 | command2 | command3
set +o pipefail

If you are executing a fragment, and unsure about the pipefail already set, you can use this with the subshell as previously suggested:

# explicitly execute with pipefail set
(set -o pipefail ; command1 | command2 | command3 )
# explicitly execute with pipefail unset
(set +o pipefail ; command1 | command2 | command3 )