The Container.State struct holds the container's state, and most of
its fields are expected to change dynamically. Some o these state-changes
are explicit, for example, setting the container to be "stopped". Other
state changes can be more explicit, for example due to the containers'
process exiting or being "OOM" killed by the kernel.
The distinction between explicit ("desired") state changes and "state"
("actual state") is sometimes vague; for some properties, we clearly
separated them, for example if a user requested the container to be
stopped or restarted, we store state in the Container object itself;
HasBeenManuallyStopped bool // used for unless-stopped restart policy
HasBeenManuallyRestarted bool `json:"-"` // used to distinguish restart caused by restart policy from the manual one
Other properties are more ambiguous. such as "HasBeenStartedBefore" and
"RestartCount", which are stored on the Container (and persisted to
disk), but may be more related to "actual" state, and likely should
not be persisted;
RestartCount int
HasBeenStartedBefore bool
Given that (per the above) concurrency must be taken into account, most
changes to the `container.State` struct should be protected; here's where
things get blurry. While the `State` type provides various accessor methods,
only some of them take concurrency into account; for example, [State.IsRunning]
and [State.GetPID] acquire a lock, whereas [State.ExitCodeValue] does not.
Even the (commonly used) [State.StateString] has no locking at all.
The way to handle this is error-prone; [container.State] contains a mutex,
and it's exported. Given that its embedded in the [container.Container]
struct, it's also exposed as an exported mutex for the container. The
assumption here is that by "merging" the two, the caller to acquire a lock
when either the container _or_ its state must be mutated. However, because
some methods on `container.State` handle their own locking, consumers must
be deeply familiar with the internals; if both changes to the `Container`
AND `Container.State` must be made. This gets amplified more as some
(exported!) methods, such as [container.SetRunning] mutate multiple fields,
but don't acquire a lock (so expect the caller to hold one), but their
(also exported) counterpart (e.g. [State.IsRunning]) do.
It should be clear from the above, that this needs some architectural
changes; a clearer separation between "desired" and "actual" state (opening
the potential to update the container's config without manually touching
its `State`), possibly a method to obtain a read-only copy of the current
state (for those querying state), and reviewing which fields belong where
(and should be persisted to disk, or only remain in memory).
This PR preserves the status quo; it makes no structural changes, other
than exposing where we access the container's state. Where previously the
State fields and methods were referred to as "part of the container"
(e.g. `ctr.IsRunning()` or `ctr.Running`), we now explicitly reference
the embedded `State` (`ctr.State.IsRunning`, `ctr.State.Running`).
The exception (for now) is the mutex, which is still referenced through
the embedded struct (`ctr.Lock()` instead of `ctr.State.Lock()`), as this
is (mostly) by design to protect the container, and what's in it (including
its `State`).
[State.IsRunning]: c4afa77157/daemon/container/state.go (L205-L209)
[State.GetPID]: c4afa77157/daemon/container/state.go (L211-L216)
[State.ExitCodeValue]: c4afa77157/daemon/container/state.go (L218-L228)
[State.StateString]: c4afa77157/daemon/container/state.go (L102-L131)
[container.State]: c4afa77157/daemon/container/state.go (L15-L23)
[container.Container]: c4afa77157/daemon/container/container.go (L67-L75)
[container.SetRunning]: c4afa77157/daemon/container/state.go (L230-L277)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
These comments were added to enforce using the correct import path for
our packages ("github.com/docker/docker", not "github.com/moby/moby").
However, when working in go module mode (not GOPATH / vendor), they have
no effect, so their impact is limited.
Remove these imports in preparation of migrating our code to become an
actual go module.
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Define consts for the Actions we use for events, instead of "ad-hoc" strings.
Having these consts makes it easier to find where specific events are triggered,
makes the events less error-prone, and allows documenting each Action (if needed).
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
The containerd client is very chatty at the best of times. Because the
libcontained API is stateless and references containers and processes by
string ID for every method call, the implementation is essentially
forced to use the containerd client in a way which amplifies the number
of redundant RPCs invoked to perform any operation. The libcontainerd
remote implementation has to reload the containerd container, task
and/or process metadata for nearly every operation. This in turn
amplifies the number of context switches between dockerd and containerd
to perform any container operation or handle a containerd event,
increasing the load on the system which could otherwise be allocated to
workloads.
Overhaul the libcontainerd interface to reduce the impedance mismatch
with the containerd client so that the containerd client can be used
more efficiently. Split the API out into container, task and process
interfaces which the consumer is expected to retain so that
libcontainerd can retain state---especially the analogous containerd
client objects---without having to manage any state-store inside the
libcontainerd client.
Signed-off-by: Cory Snider <csnider@mirantis.com>
If "ps" fails, in many cases it prints a meaningful error message
which a user can benefit from. Let's use it.
While at it, let's use errdefs.System to classify the error,
as well as errors.Wrap.
Before:
> $ docker top $CT <any bad ps options>
> Error response from daemon: Error running ps: exit status 1
After:
> $ docker top $CT auxm
> Error response from daemon: ps: error: thread display conflicts with forest display
or
> $ docker top $CT saur
> Error response from daemon: ps: error: conflicting format options
or, if there's no meaningful error on stderr, same as before:
> $ docker top $CT 1234
> Error response from daemon: ps: exit status 1
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Current ContainerTop (a.k.a. docker top) implementation uses "ps"
to get the info about *all* running processes, then parses it, then
filters the results to only contain PIDs used by the container.
Collecting data only to throw most of it away is inefficient,
especially on a system running many containers (or processes).
For example, "docker top" on a container with a single process
can take up to 0.5 seconds to execute (on a mostly idle system)
which is noticeably slow.
Since the containers PIDs are known beforehand, let's use ps's
"-q" option to provide it with a list of PIDs we want info about.
The problem with this approach is, some ps options can't be used
with "-q" (the only one I'm aware of is "f" ("forest view") but
there might be more). As the list of such options is not known,
in case ps fails, it is executed again without "q" (retaining
the old behavior).
Next, the data produced by "ps" is filtered in the same way as before.
The difference here is, in case "-q" worked, the list is much shorter.
I ran some benchmarks on my laptop, with about 8000 "sleep" processes
running to amplify the savings.
The improvement in "docker top" execution times is 5x to 10x (roughly
0.05s vs 0.5s).
The improvement in ContainerTop() execution time is up to 100x
(roughly 3ms vs 300ms).
I haven't measured the memory or the CPU time savings, guess those
are not that critical.
NOTE that busybox ps does not implement -q so the fallback is always
used, but AFAIK it is not usable anyway and Docker expects a normal
ps to be on the system (say the list of fields produced by
"busybox ps -ef" differs from normal "ps -ef" etc.).
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
It does not make sense to keep looking for PID once
we found it, so let's give it a break.
The side effect of this patch is, if there's more than one column
titled "PID", the last (rightmost) column was used before, and now
the first (leftmost) column is used. Should make no practical
difference whatsoever.
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
Use strongly typed errors to set HTTP status codes.
Error interfaces are defined in the api/errors package and errors
returned from controllers are checked against these interfaces.
Errors can be wraeped in a pkg/errors.Causer, as long as somewhere in the
line of causes one of the interfaces is implemented. The special error
interfaces take precedence over Causer, meaning if both Causer and one
of the new error interfaces are implemented, the Causer is not
traversed.
Signed-off-by: Brian Goff <cpuguy83@gmail.com>
Changed per requested review to refactor to make it more logic clear.
Current output for "docker top <contianer-id> m" option,eg:
root@b2c7ec58399d:/go/src/github.com/docker/docker# docker top 755d5871ec45 am
PID TTY STAT TIME COMMAND
148 pts/0 - 0:00 bash
- - Ss+ 0:00 -
fixing issue:#30580
Signed-off-by: catinthesky <yaozaiyong@hotmail.com>
Moving all strings to the errors package wasn't a good idea after all.
Our custom implementation of Go errors predates everything that's nice
and good about working with errors in Go. Take as an example what we
have to do to get an error message:
```go
func GetErrorMessage(err error) string {
switch err.(type) {
case errcode.Error:
e, _ := err.(errcode.Error)
return e.Message
case errcode.ErrorCode:
ec, _ := err.(errcode.ErrorCode)
return ec.Message()
default:
return err.Error()
}
}
```
This goes against every good practice for Go development. The language already provides a simple, intuitive and standard way to get error messages, that is calling the `Error()` method from an error. Reinventing the error interface is a mistake.
Our custom implementation also makes very hard to reason about errors, another nice thing about Go. I found several (>10) error declarations that we don't use anywhere. This is a clear sign about how little we know about the errors we return. I also found several error usages where the number of arguments was different than the parameters declared in the error, another clear example of how difficult is to reason about errors.
Moreover, our custom implementation didn't really make easier for people to return custom HTTP status code depending on the errors. Again, it's hard to reason about when to set custom codes and how. Take an example what we have to do to extract the message and status code from an error before returning a response from the API:
```go
switch err.(type) {
case errcode.ErrorCode:
daError, _ := err.(errcode.ErrorCode)
statusCode = daError.Descriptor().HTTPStatusCode
errMsg = daError.Message()
case errcode.Error:
// For reference, if you're looking for a particular error
// then you can do something like :
// import ( derr "github.com/docker/docker/errors" )
// if daError.ErrorCode() == derr.ErrorCodeNoSuchContainer { ... }
daError, _ := err.(errcode.Error)
statusCode = daError.ErrorCode().Descriptor().HTTPStatusCode
errMsg = daError.Message
default:
// This part of will be removed once we've
// converted everything over to use the errcode package
// FIXME: this is brittle and should not be necessary.
// If we need to differentiate between different possible error types,
// we should create appropriate error types with clearly defined meaning
errStr := strings.ToLower(err.Error())
for keyword, status := range map[string]int{
"not found": http.StatusNotFound,
"no such": http.StatusNotFound,
"bad parameter": http.StatusBadRequest,
"conflict": http.StatusConflict,
"impossible": http.StatusNotAcceptable,
"wrong login/password": http.StatusUnauthorized,
"hasn't been activated": http.StatusForbidden,
} {
if strings.Contains(errStr, keyword) {
statusCode = status
break
}
}
}
```
You can notice two things in that code:
1. We have to explain how errors work, because our implementation goes against how easy to use Go errors are.
2. At no moment we arrived to remove that `switch` statement that was the original reason to use our custom implementation.
This change removes all our status errors from the errors package and puts them back in their specific contexts.
IT puts the messages back with their contexts. That way, we know right away when errors used and how to generate their messages.
It uses custom interfaces to reason about errors. Errors that need to response with a custom status code MUST implementent this simple interface:
```go
type errorWithStatus interface {
HTTPErrorStatusCode() int
}
```
This interface is very straightforward to implement. It also preserves Go errors real behavior, getting the message is as simple as using the `Error()` method.
I included helper functions to generate errors that use custom status code in `errors/errors.go`.
By doing this, we remove the hard dependency we have eeverywhere to our custom errors package. Yes, you can use it as a helper to generate error, but it's still very easy to generate errors without it.
Please, read this fantastic blog post about errors in Go: http://dave.cheney.net/2014/12/24/inspecting-errors
Signed-off-by: David Calavera <david.calavera@gmail.com>
Although having a request ID available throughout the codebase is very
valuable, the impact of requiring a Context as an argument to every
function in the codepath of an API request, is too significant and was
not properly understood at the time of the review.
Furthermore, mixing API-layer code with non-API-layer code makes the
latter usable only by API-layer code (one that has a notion of Context).
This reverts commit de41640435, reversing
changes made to 7daeecd42d.
Signed-off-by: Tibor Vass <tibor@docker.com>
Conflicts:
api/server/container.go
builder/internals.go
daemon/container_unix.go
daemon/create.go
This reverts commit ff92f45be4, reversing
changes made to 80e31df3b6.
Reverting to make the next revert easier.
Signed-off-by: Tibor Vass <tibor@docker.com>
Avoid creating a global context object that will be used while the daemon is running.
Not only this object won't ever be garbage collected, but it won't ever be used for anything else than creating other contexts in each request. I think it's a bad practive to have something like this sprawling aroud the code.
This change removes that global object and initializes a context in the cases we don't have already one, like shutting down the server.
This also removes a bunch of context arguments from functions that did nothing with it.
Signed-off-by: David Calavera <david.calavera@gmail.com>
This PR adds a "request ID" to each event generated, the 'docker events'
stream now looks like this:
```
2015-09-10T15:02:50.000000000-07:00 [reqid: c01e3534ddca] de7c5d4ca927253cf4e978ee9c4545161e406e9b5a14617efb52c658b249174a: (from ubuntu) create
```
Note the `[reqID: c01e3534ddca]` part, that's new.
Each HTTP request will generate its own unique ID. So, if you do a
`docker build` you'll see a series of events all with the same reqID.
This allow for log processing tools to determine which events are all related
to the same http request.
I didn't propigate the context to all possible funcs in the daemon,
I decided to just do the ones that needed it in order to get the reqID
into the events. I'd like to have people review this direction first, and
if we're ok with it then I'll make sure we're consistent about when
we pass around the context - IOW, make sure that all funcs at the same level
have a context passed in even if they don't call the log funcs - this will
ensure we're consistent w/o passing it around for all calls unnecessarily.
ping @icecrime @calavera @crosbymichael
Signed-off-by: Doug Davis <dug@us.ibm.com>
- some method names were changed to have a 'Locking' suffix, as the
downcased versions already existed, and the existing functions simply
had locks around the already downcased version.
- deleting unused functions
- package comment
- magic numbers replaced by golang constants
- comments all over
Signed-off-by: Morgan Bauer <mbauer@us.ibm.com>
