SIGCHLD and SIGWINCH used in api/client (cli code) are not
available on Windows. Extracting into separate files with build
tags.
Signed-off-by: Ahmet Alp Balkan <ahmetb@microsoft.com>
Signal proxy does work only in non-TTY mode (--tty=false). Man pages and
commands should not lie about it.
Signed-off-by: Michal Minar <miminar@redhat.com>
Current implementation is hard to reason about because of trying to mix
unix/tcp server implementations, even though they are quite different.
This cleans that up.
Also makes it possible to create and manage a new API server easily,
e.g. for adding an introspection socket to a container.
Built in such a way as to allow a non-HTTP server to work as well, such
as libchan.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
This patch fixes the compilation errors in Docker due to changes in the
libcontainer/user API. There is no functionality change due to this
patch.
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com> (github: cyphar)
E.g. "docker foobar run" would have printed "Command not found:
foobar" and printed the help text for "run". It should instead
print the root help message for docker.
Signed-off-by: Ben Firshman <ben@firshman.co.uk>
While working on the fix for #8330 I noticed a few things:
1 - the split() call for the .dockerignore process will generate a blank
"exclude". While this isn't causing an issue right now, I got worried
that in the future some code later on might interpret "" as something bad,
like "everything" or ".". So I added a check for an empty "exclude"
and skipped it
2 - if someone puts "foo" in their .dockerignore then we'll skip "foo".
However, if they put "./foo" then we won't due to the painfully
simplistic logic of go's filepath.Match algorithm. To help things
a little (and to treat ./Dockerfile just like Dockerfile) I added
code to filepath.Clean() each entry in .dockerignore. It should
result in the same semantic path but ensure that no matter how the
user expresses the path, we'll match it.
Signed-off-by: Doug Davis <dug@us.ibm.com>
With current implementation there was a possibility
that /start responds quicker than /attach, meaning that
some output would be clipped.
Fixed so the implementation matches with `docker run`.
This also fixes the flaky test results for TestCreateEchoStdout.
Signed-off-by: Tõnis Tiigi <tonistiigi@gmail.com> (github: tonistiigi)
This will allow us to use a common Git prefix check for both api/clients/commands.go and
builder/job.go. Previous prefix check in build from Git (in builder/jobs.go) ignored valid prefixes such as "git@", "http://" or "https://".
Signed-off-by: Lakshan Perera <lakshan@laktek.com>
First off, sorry for the noise. This is a cleaner step of #8508
Found more of a root cause of the open file handles.
After more testing I found that the open file descriptors will still
occur for TCP:// connections to the daemon, causing client and/or daemon
to fail.
The issue was instantiating a new http.Transport on _ever_ client
request. So each instance held the prior connection alive, but was only
ever used once.
By moving it out to the initilization of DockerCli, we can now have
reuse of idled connections. Simplifies the garbage overhead of the
client too, though that's not usually a deal.
Signed-off-by: Vincent Batts <vbatts@redhat.com>
In the go stdlib net/http Transport, the used connections are cached
when idled. This behaviour is intended for TCP connections and does not
behave correctly for unix sockets. Despite the
DefaultMaxIdleConnsPerHost being 2, the idled connections are held open
during a session. For large sessions like `docker rm $(docker ps -a -q)`
of thousands of containers, it will cause the client _and_ the server to
open too many fails and have failures.
Having keep alives not used for only unix sockets is a work around for
this stdlib issue.
Also this includes disabling compression when communicating over the
local unix socket too.
Signed-off-by: Vincent Batts <vbatts@redhat.com>
