New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Successive docker starts do not start container every time #4410
Comments
For at least the case where we get a 125 and no output, I found out that the start command does not start the container. We get the powered off event for the previous start call late, and set the container state to stopped for the current call before we can actually power on the VM, which results in the container not getting started. |
Where do we abort a start call based on the container state? It's supposed to be an idempotent operation, so portlayer would just do nothing if the container is already in the desired state - but that should be being determined based on the VM runtime structure, not the container cache state, which is primarily for faking the transition states. |
@hmahmood Please prioritize and estimate and move to backlog. |
vspc holds the connection even after the vm powers off up to 10 sec. so starting the vm again before this loops run causes issues.
|
@mhagen-vmware should this go in the Release Notes for 1.1? |
* Attach support for exec'ed session Includes @hmahmood's and @hickeng's various fixed to dio/persona and portlayer Fixes #4263 Fixes #4367 Fixes #4598 Fixes #4410 Fixes #4339 Fixes #4881 Requires #4288 Implements exec functionality for personality Those changes live in lib/apiservers/engine/backends/ directory. There are changes in container_proxy functions since exec and attach shares some common code paths. We started to pass ids instead of viccontainer structs around to them to support both workflows. AttachStreams function splits the cancelation logic into different parts. Closing stdin cancels other active streams. We also ensure that the API client connection is shutdown on exit. Implements exec support for portlayer rest api Those changes live in lib/apiservers/portlayer/ directory. One important change there is context changes. In the past we were calculating the timeouts manually and then passing timeouts as a parameter to Get call. session, err := attachServer.Get(context.Background(), params.ID, timeout) now each stream is creating its own context and passing it to the function - eliminating manual calculations and redundant timeout value. session, err := i.server.Interaction(ctx, params.ID) Portlayer attach subsystem re-organization Those changes live in lib/portlayer/attach/. Some of them just renames and splitting monolithic file (they are now bind.go/join.go etc.) into pieces. Connector has new set of features. We now have a different unblock signal for signaling tether (called unblock) for starting the processes inside the container. We also now have a SSH channel mux running on client side to receive container ids from tether. Normally portlayer initiates the connection and asks container ids. This is still the case for the primary session. But after that primary session start event, any exec'ed session will cause tether to restart and as a part of that tether will push container ids to the portlayer. Additionally we now started to use channel mux to evict closed connections from the connection map (#4367). Lastly we started to use singleflight to make sure that there is only one call in-flight for each ID at any given time. tether support for exec As noted above, tether now pushes container ids to portlayer when it receives reload signal. It supports new unblock signal. Fixes a bug caused by wrong usage of cleanup functions. Fixes races causing incorrect behaviors. dio changes dio reader now has a PropagateEOF function toggles whether to return EOF when all readers return EOF. Setting this to true will result in an EOF if there are no readers available when Read is next called. dio writer now calls CloseWrite if the writes implements CloseWriter.
* Attach support for exec'ed session Includes @hmahmood's and @hickeng's various fixed to dio/persona and portlayer Fixes vmware#4263 Fixes vmware#4367 Fixes vmware#4598 Fixes vmware#4410 Fixes vmware#4339 Fixes vmware#4881 Requires vmware#4288 Implements exec functionality for personality Those changes live in lib/apiservers/engine/backends/ directory. There are changes in container_proxy functions since exec and attach shares some common code paths. We started to pass ids instead of viccontainer structs around to them to support both workflows. AttachStreams function splits the cancelation logic into different parts. Closing stdin cancels other active streams. We also ensure that the API client connection is shutdown on exit. Implements exec support for portlayer rest api Those changes live in lib/apiservers/portlayer/ directory. One important change there is context changes. In the past we were calculating the timeouts manually and then passing timeouts as a parameter to Get call. session, err := attachServer.Get(context.Background(), params.ID, timeout) now each stream is creating its own context and passing it to the function - eliminating manual calculations and redundant timeout value. session, err := i.server.Interaction(ctx, params.ID) Portlayer attach subsystem re-organization Those changes live in lib/portlayer/attach/. Some of them just renames and splitting monolithic file (they are now bind.go/join.go etc.) into pieces. Connector has new set of features. We now have a different unblock signal for signaling tether (called unblock) for starting the processes inside the container. We also now have a SSH channel mux running on client side to receive container ids from tether. Normally portlayer initiates the connection and asks container ids. This is still the case for the primary session. But after that primary session start event, any exec'ed session will cause tether to restart and as a part of that tether will push container ids to the portlayer. Additionally we now started to use channel mux to evict closed connections from the connection map (vmware#4367). Lastly we started to use singleflight to make sure that there is only one call in-flight for each ID at any given time. tether support for exec As noted above, tether now pushes container ids to portlayer when it receives reload signal. It supports new unblock signal. Fixes a bug caused by wrong usage of cleanup functions. Fixes races causing incorrect behaviors. dio changes dio reader now has a PropagateEOF function toggles whether to return EOF when all readers return EOF. Setting this to true will result in an EOF if there are no readers available when Read is next called. dio writer now calls CloseWrite if the writes implements CloseWriter.
* Attach support for exec'ed session Includes @hmahmood's and @hickeng's various fixed to dio/persona and portlayer Fixes vmware#4263 Fixes vmware#4367 Fixes vmware#4598 Fixes vmware#4410 Fixes vmware#4339 Fixes vmware#4881 Requires vmware#4288 Implements exec functionality for personality Those changes live in lib/apiservers/engine/backends/ directory. There are changes in container_proxy functions since exec and attach shares some common code paths. We started to pass ids instead of viccontainer structs around to them to support both workflows. AttachStreams function splits the cancelation logic into different parts. Closing stdin cancels other active streams. We also ensure that the API client connection is shutdown on exit. Implements exec support for portlayer rest api Those changes live in lib/apiservers/portlayer/ directory. One important change there is context changes. In the past we were calculating the timeouts manually and then passing timeouts as a parameter to Get call. session, err := attachServer.Get(context.Background(), params.ID, timeout) now each stream is creating its own context and passing it to the function - eliminating manual calculations and redundant timeout value. session, err := i.server.Interaction(ctx, params.ID) Portlayer attach subsystem re-organization Those changes live in lib/portlayer/attach/. Some of them just renames and splitting monolithic file (they are now bind.go/join.go etc.) into pieces. Connector has new set of features. We now have a different unblock signal for signaling tether (called unblock) for starting the processes inside the container. We also now have a SSH channel mux running on client side to receive container ids from tether. Normally portlayer initiates the connection and asks container ids. This is still the case for the primary session. But after that primary session start event, any exec'ed session will cause tether to restart and as a part of that tether will push container ids to the portlayer. Additionally we now started to use channel mux to evict closed connections from the connection map (vmware#4367). Lastly we started to use singleflight to make sure that there is only one call in-flight for each ID at any given time. tether support for exec As noted above, tether now pushes container ids to portlayer when it receives reload signal. It supports new unblock signal. Fixes a bug caused by wrong usage of cleanup functions. Fixes races causing incorrect behaviors. dio changes dio reader now has a PropagateEOF function toggles whether to return EOF when all readers return EOF. Setting this to true will result in an EOF if there are no readers available when Read is next called. dio writer now calls CloseWrite if the writes implements CloseWriter.
One docker start call after another in quick succession on the same container sometimes will return 125 exit code from the docker cli.
Steps to reproduce:
docker create --name test busybox echo foo
docker start -a test
; should return "foo" and exit code 0 every time, but sometimes will return 125 with or without the correct outputThe text was updated successfully, but these errors were encountered: