Adding keys tells Ember to rerender matching entries instead of
destroying and recreating.
Without this key, every time the allocation collection changes, every
allocation row gets destroyed and recreated.
This happens a lot, since each allocation needs to be reloaded which
dirties the collection.
Since allocation rows fetch stats on init, each of these many many
renders results in a stats request.
By using key and rerendering matching records, this all goes away. Since
the rows aren't being destroyed and recreated, the init stats request
isn't being made overnumerously.
This introduces a DataCaches service so recently-updated collections don’t need
to be requeried within a minute, or based on the current route. It only searches
jobs and nodes. There are known bugs that will be addressed in upcoming PRs.
This was a disturbing discovery. Requests in watch loops would recycle
AbortControllers meaning once any request was aborted, all requests
forever after were skipped. I noticed it with deployments and job
summary on the job detail page.
I suspect this regression occurred when jQuery was removed. This needs
test coverage still to make sure it doesn't happen again.
The id-processing in the WatchableNamespaceIds adapter was
happening twice; this removes urlForUpdate record so it
only happens once. @DingoEatingFuzz figured it out! 🥳
Originally this was failing because it only had a getter.
I tried replacing it with a computed property and that
succeeded, but since we have already stopped using
jQuery, we might as well remove it.
Manual interventions:
• decorators on the same line for service and controller
injections and most computed property macros
• preserving import order when possible, both per-line
and intra-line
• moving new imports to the bottom
• removal of classic decorator for trivial cases
• conversion of init to constructor when appropriate
This is extracted from #8094, where I have run into some snags. Since
these ESLint fixes aren’t actually connected to the Ember 3.16 update
but involve changes to many files, we might as well address them
separately. Where possible I fixed the problems but in cases where
a fix seemed too involved, I added per-line or -file exceptions.
This is mostly a direct application of the ember-angle-brackets-codemod.
I manually restored newlines in multi-line component invocations, usually
preserving file line length except for now-non-positional link-to @route.
I needed to rename task to taskState in some cases to avoid Ember
Concurrency naming conflicts.
This partially addresses #7799.
Task state filesystems are contained within a subdirectory of their
parent allocation, so almost everything that existed for browsing task
state filesystems was applicable to browsing allocations, just without
the task name prepended to the path. I aimed to push this differential
handling into as few contained places as possible.
The tests also have significant overlap, so this includes an extracted
behavior to run the same tests for allocations and task states.
This updates Xterm.js to 4.6.0, which includes support for reverse-wraparound
mode, so we no longer need to use a vendored dependency, which closes#7461.
The interface for accessing the buffer that’s used for test assertions changed.
With the dependency now accessed conventionally, we can have it load only when
it’s needed by an exec popup window, which closes#7516. That saves us
≈60kb compressed in the dependency bundle!
Going off of the error message being "Forbidden" was brittle to begin
with and no longer works with Fetch due to the error message coming from
jquery underpinnings that were unobserved by Ember Data's attempted
recreation.
Typically these filterable list views don't have titles beyond the
breadcrumbs, but since this page has no search bar, the title really
helps balance it out.
This fixes a bug in #7815 where you can’t open an exec window from
the allocation overview because accessing `allocation.job.plainId`
fails across the proxied relationship.
Changing namespaces can be done anywhere in the app even though many
Nomad resources aren't namespace-sensitive (e.g., clients, plugins).
A user changing namespaces is an intent to reset context, "now I want
to begin a task that relates to Namespace X". Where that task begins
used to always be the Jobs list, since it was the only namespace
sensitive resource. Now with CSI Volumes, "square 1" is Volumes if the
namespace is changed from a storage page.
There was a missing edge case where a job is pending. I took the moment
to also refactor the code to use async/await which cleaned up the
promise chaining.
Structure icons have fill set to currentColor hardcored in their markup.
This mean setting fill to a color in CSS does nothing, but setting color
now does.
In the client view list, only show running allocations count for each
client, rather than include already completed tasks.
This is done for two reasons:
First, consitency with the CLI: `nomad node status --allocs` only
shows running allocs.
Second, and more importantly, the count is useful to estimate how loaded
the clients are. Allocs that have completed (but not GCed yet) have
very little value to operators.
This would happen because a no connection error happens after the second request fails, but
that's because it's assumed the second request is to a server node. However, if a user clicks
stderr fast enough, the first and second requests are both to the client node. This changes
the logic to check if the request is to the server before deeming log streaming a total failure.
Typically a failover means that the client can't be reached. However, if
the client does eventually return after the timeout period, the log will
stream indefinitely. This fixes that using an API that wasn't broadly
available at the time this was first written.
This closes#7456. It hides the terminal when the job is dead and
displays an error when trying to open an exec session for a task
that isn’t running. There’s a skipped test for the latter behaviour
that I’ll have to come back for.
This closes#7454. It makes use of the existing watchable tools to
allow the exec popup sidebar to be live-updating. It also adds
alphabetic sorting of task groups and tasks.
This is a minimal implementation that closes#7463. It doesn’t include
true support for moving around within the command to edit using arrow
keys because it gets too complex when managing wrapping at the edge of
the terminal. Instead, arrow keys are ignored. It also ignores ^A and
^E, which are cursor manipulations that pose similar problems to arrow
keys. It does support ^U, which deletes the entire command.
It also allows a command to be pasted, which was previously unsupported.
This is accomplished by migrating from Xterm.js’s onKey handler to
onData, which is recommended here:
https://github.com/xtermjs/xterm.js/issues/2673#issuecomment-574897733
onData is a higher-level handler that issues events with the final
interpreted data instead of the individual key events. That means the
processing in this PR has changed from inspecting DOM key events to
inspecting their ASCII equivalents, which I’ve extracted into a utility
dictionary for use in tests and implementation.
One consequence of ignoring most control characters is that if you paste
a string that includes a control character, that character will be
stripped. It’s somewhat strange for compound sequences like arrow keys;
if you run copy('/bin/b' + '\x1b[D' + 'ash') in a Javascript console and
paste what’s on the clipboard, you get "/bin/b[Dash". That’s because
the left arrow key, as in that centre portion of the string,
is represented by the escape character and a coded sequence. Stripping
the control character leaves the coded sequence as part of the paste.
That seems like an acceptable compromise vs either ignoring any pasted
string with control characters (confusing UX) or trying to interpret and
strip all such compound control sequences (difficult to be exhaustive).
This effectively overrides Bulma's default field layout tweaks
at different breakpoints. This includes going from flex to block
and different font-sizes.
The "default" order values as set by Bulma are different for different
breakpoints. Since this wasn't considering breakpoints, it resulted
in the unexpected reordering of pagination elements as different page
widths. Turns out removing this property gives us what we want.
When a node has no host volumes, the API response will
have a null value for the HostVolumes attribute, which
in turn becomes a null value instead of an empty array
in the store. This protects against that, ensuring host
volumes is always an array.
This closes#7476. The decomposition of computed properties
is necessary to avoid nested aggregate dependent keys; the
previous dependent key of `taskGroup` will be inadequate
when the sidebar becomes live-updating.
Closes#7197#7199
Note: Test coverage is limited to adapter and serializer unit tests. All
acceptance tests have been stubbed and all features have been manually
tested end-to-end.
This represents Phase 1 of #6993 which is the core workflow of CSI in
the UI. It includes a couple new pages for viewing all external volumes
as well as the allocations associated with each. It also updates
existing volume related views on job and allocation pages to handle both
Host Volumes and CSI Volumes.
This connects Xterm.js to a Nomad exec websocket so people
can interact on clients via live sessions. There are buttons on
job, allocation, task group, and task detail pages that open a
popup that lets them edit their shell command and start a
session.
More is to come, as recorded in issues.
I originally planned to add component documentation, but as this dragged on and I found that JSDoc-to-Markdown sometimes needed hand-tuning, I decided to skip it and focus on replicating what was already present in Freestyle. Adding documentation is a finite task that can be revisited in the future.
My goal was to migrate everything from Freestyle with as few changes as possible. Some adaptations that I found necessary:
• the DelayedArray and DelayedTruth utilities that delay component rendering until slightly after initial render because without them:
◦ charts were rendering with zero width
◦ the JSON viewer was rendering with empty content
• Storybook in Ember renders components in a routerless/controllerless context by default, so some component stories needed changes:
◦ table pagination/sorting stories access to query params, which necessitates some reaching into Ember internals to start routing and dynamically generate a Storybook route/controller to render components into
◦ some stories have a faux controller as part of their Storybook context that hosts setInterval-linked dynamic computed properties
• some jiggery-pokery with anchor tags
◦ inert href='#' had to become href='javascript:;
◦ links that are actually meant to navigate need target='_parent' so they don’t navigate inside the Storybook iframe
Maybe some of these could be addressed by fixes in ember-cli-storybook but I’m wary of digging around in there any more than I already have, as I’ve lost a lot of time to Storybook confusion and frustrations already 😞
The STORYBOOK=true environment variable tweaks some environment settings to get things working as expected in the Storybook context.
I chose to:
• use angle bracket invocation within stories rather than have to migrate them soon after having moved to Storybook
• keep Freestyle around for now for its palette and typeface components
This builds on API changes in #6017 and #6021 to conditionally turn off the
“Run Job” button based on the current token’s capabilities, or the capabilities
of the anonymous policy if no token is present.
If you try to visit the job-run route directly, it redirects to the job list.
There are two changes here, and some caveats/commentary:
1. The “State“ table column was actually sorting only by status. The state was not an actual property, just something calculated in each client row, as a product of status, isEligible, and isDraining. This PR adds isDraining as a component of compositeState so it can be used for sorting.
2. The Sortable mixin declares dependent keys that cause the sort to be live-updating, but only if the members of the array change, such as if a new client is added, but not if any of the sortable properties change. This PR adds a SortableFactory function that generates a mixin whose listSorted computed property includes dependent keys for the sortable properties, so the table will live-update if any of the sortable properties change, not just the array members. There’s a warning if you use SortableFactory without dependent keys and via the original Sortable interface, so we can eventually migrate away from it.
As the angle bracket invocation RFC says:
> There is no dedicated syntax for passing an "else" block
> directly. If needed, that can be passed using the named
> blocks syntax.
https://github.com/emberjs/rfcs/blob/master/text/0311-angle-bracket-invocation.md#block
Unfortunately, using a contextual component doesn’t help as
the yield inside that component will still result in content
rendering that would show when the source isn’t empty. So
we decided to change the interface so you have to check
whether the source is empty before using it, which aligns with
how list-table works.
This is mostly deprecation fixes and blueprint changes. There
are some dependency updates too; the changes to Ember
Basic Dropdown necessitated changing it to angle bracket
component invocation. The conversion of the rest of the
templates will happen separately.
This fixes a race condition in the pseudo-relationship between a
TaskState and a Task that was causing the Consul Connect proxy tag
to sometimes show on the wrong task. There’s no direct Ember Data-style
relationship between a TaskState and its Task; instead, it’s determined
by searching for a Task with the matching name. The related Task was
sometimes stored before everything was ready and not recalculated when
the name became known. This ensures the relationship is accurate if the
TaskState’s name property changes.
Protect against case where an alloc has no services and avoid
dereferencing null.
Here, we ensure that the model and test serializers mimic the API by
having nil TaskGroup.Services instead of an empty array.
I put this property in the wrong place.
I’ve found how to fix the mock API in the tests but
they’re failing to pass with headless Chrome only,
so they’re skipped for now.
This was causing elements to flow off the page, since the element was
assuming 100% but also had a 250px margin for the left column.
This had previously been "fixed" by setting overflow-x: auto, but that
resulted in tooltips from being clipped.
This is a better solution to the same problem.
It may be an Ember bug: in some circumstances, the
ember-transitioning-in class was persisting in table
sort links even after the transition completed. This
changes the transition animations to be targeted only
for breadcrumbs and directory links.
The new streaming-file component takes an arbitrary logger component
along with some mode flags and handles things like polling, DOM updates,
and scroll position.
When sorting by size, directories are sorted by name, as size
isn’t displayed.
This includes a change to the positioning of sort arrows for all tables,
moving them closer to the text, because in some cases, the arrows
for right-aligned columns were ambiguously positioned.
This uses ember-page-title to add dynamic page titles throughout the
route hierarchy. When there’s more than one region, the current
current region is added before the final entry of “- Nomad”.
* origin/master: (32 commits)
Added additional test cases and fixed go test case
update changelog
Add Mirage-toggling via environment variable (#5899)
changelog: Add entries for windows fixes
fifo: Safer access to Conn
run post-run/post-stop task runner hooks
Fail alloc if alloc runner prestart hooks fail
address review comments
changelog
Missed one revert of backwards compatibility for node drain
Improve test cases for detecting content type
Undo removal of node drain compat changes
Updated with suggestions.
fifo: Close connections and cleanup lock handling
logmon: Add windows compatibility test
client: defensive against getting stale alloc updates
Infer content type in alloc fs stat endpoint
appveyor: Run logmon tests
fifo: Require that fifos do not exist for create
vendor: Use dani fork of go-winio
...
The draining, eligibility, and status fields now all show under a combined
state column. Draining takes precedence, then (in)eligibility; if neither of
those is true, the status displays.
This changes the templates so the element that contains
the search box is always present, instead hiding only
the box itself when there’s nothing to search. Keeping
the empty element lets it take up its flexbox space so
the facets will no longer be in the centre.
1. Label no longer bleeds into the downward arrow icon
2. The selection number no longer breaks onto its own line
3. Options have a min-width so short labels can't result in small click
targets
The impact was the application error was no longer being nulled out,
causing the application error to continue to be shown after
transitioning.
This never happened in apps since it's not possible to transition away
from the error screen.
Since DOM code is in a run.next, it's possible that between the DOM
code being queued and running the element is destroyed. So the DOM
code needs to guard against this using the isDestroyed API.
1. Check if the response is a 4xx/5xx
2. If it is, skip the append step and track a frame miss
3. If enough frame misses occur in a row, treat it as a pause
A "pause" is when a null data frame is added, which shows up as a
gap in line charts.
Searchable can be used without pagination, but reseting pagination
is more a function of search than pagination insofar as if you
add search to a page, you are also going to want automatic pagination
resetting.
This is the best of three options
1. Users of stats trackers control polling (old method)
2. Stat tracker is stateful and has start/stop methods (like logging)
3. Stat trackers blindly throttle requests
This is the best option because it means N number of concurrent users of
a stats tracker can request polling without inundating the tracker with
redundant frames (or the network with redundant requests), but they also
don't have to coordinate amongst themselves to determine what state a
tracker should be in.
This solves two problems:
1. redundant trackers making redundant requests
2. trackers being obliterated as soon as the primary metric component
is destroyed
It introduces a new problem where visiting more and more node and
allocation pages adds to an ever-growing list of trackers that can
assume lots of memory, but it solves the problem by using a
least-recently-used cache to limit the number of trackers tracked.
When data is coming in live, the tooltip can get bogged down by updates
causing the tooltip to never make it under the mouse, which looks like
either lag or a bug.
