| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| | |
|
| |
|
|
| |
This line is too long.
|
| |
|
|
|
| |
We already log the outcome of the HsDir upload in the function that
calls this code.
|
| |
|
|
| |
This also wraps the line.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, it was possible for the `IptPublishSet` used to generate the
descriptor and the `IptPublishSet` `note_publication_attempt` to
differ. Now, the publisher generates the descriptor using the same
`IptPublishSet` it calls `note_publication_attempt` on.
Note that as a consequence, the publisher generates a new descriptor
just before _each_ HsDir upload. This means each HsDir could, in theory,
receive a different descriptor (not just in terms of revision-counters,
but also with a different set of IPTs). It may seem like this could lead
to some HsDirs being left with an outdated descriptor, but that's not
the case: after the upload completes, the publisher will be notified by
the ipt_watcher of the IPT change event (if there was one to begin
with), which will trigger another upload job.
Previously, the publisher would only generate a single descriptor for
each time period (all HsDirs in a given time period would receive the
same descriptor).
Closes #1097
|
| |
|
|
|
|
| |
This was previously in `TimePeriodUploadResult`. We will soon have
different revision_counter for each `HsDirUploadResult`, so let's
preemptively move the field there.
|
| | |
|
| |
|
|
| |
We're going to need to clone it soon.
|
| |
|
|
|
|
|
|
| |
`generate_revision_counter` and `create_ope_key` don't use anything from
`self` other than `imm`, so they might as well be methods on
`Immutable`. This change is needed because we're soon going to need to
use `generate_revision_counter` from an associated `Reactor` function
(where we don't have `self`).
|
| |\
| |
| |
| |
| |
| |
| | |
tor-hsservice: Remove a publisher TODO that has been addressed.
Closes #1131
See merge request tpo/core/arti!1807
|
| | | |
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The publisher doesn't reupload unless explicitly asked to do so by the
`IptManager` (via `await_update()`).
Also, when the consensus changes, we always trigger a reupload, but only
to those HsDirs that don't already have the descriptor (the HsDirs
marked as "clean" stay "clean", and any new HsDirs are marked "dirty"
until they get a copy of the descriptor. See !1806).
Similarly, a config change only triggers a reupload if the change means
we need to generate a new descriptor (e.g. if the `anonymity` of the
service changes). Note, however, that this logic is currently commented
out (it depends on #1028).
Closes #1131
|
| | |
| |
| |
| | |
Part of #1083
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
We need to know the status of each component to be able to report the
overall status of the service. Without this change, the service (and its
components) have no way of knowing if a given transition is valid: if
the state of a component (say, the IPT manager) is `Bootstrapping`,
`Recovering` or `Broken`, a transition out of the current state is only
valid if it is initiated by the same component that caused the current
state (for example, if the publisher sets the state to `Recovering`, the
IPT manager should not be allowed to trigger an overall state transition
to `Running`).
Part of #1083
|
| |/ |
|
| |
|
|
| |
Fixes a clippy lint.
|
| | |
|
| |
|
|
|
|
|
| |
This fixes a bug where the publisher wasn't preserving the
`DescriptorStatus` of its `HsDirs` when handling consensus changes.
The bug is described in more detailed in the TODO removed by this commit.
|
| | |
|
| |
|
|
| |
This is actually a `TimePeriodContext`.
|
| |\
| |
| |
| |
| |
| |
| | |
Improve KeySpecifier, errors, IptKeySpecifier impl, etc.
Closes #1116 and #1148
See merge request tpo/core/arti!1796
|
| | | |
|
| | | |
|
| | | |
|
| | |
| |
| |
| |
| | |
The macro is going to want to be able to do a back-conversion from an
ArtiPath, so there has to be an owned version.
|
| | |
| |
| |
| |
| | |
Now they are in the order they appear in the ArtiPath, which is a bit
clearer.
|
| | | |
|
| | | |
|
| | | |
|
| | |
| |
| |
| | |
And implement serde that way, using SerdeStringOrTransparent.
|
| | |
| |
| |
| |
| | |
And make it useable elsewhere by not expecting the particular error
type.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This allows a principled implentation via Display. The alternative
would be to have runtime panics whenever someone uses
KeySpecifierComponentViaDisplayFromStr but the Display impl can
produce wrong syntax (that doesn't fit in an ArtiPathComponent).
Bug seems like the right error since implemnting KeySpecifierComponent
implies that every value is representable in an ArtiPathComponent.
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| | |
This is a variant in KeyPathError, which contains the actual wrong
value, and the key name. But it is also a unit struct to return from
the trait impl.
This replaces ArtiPathError::InvalidDenotator.
|
| | |
| |
| |
| |
| | |
Save the ArtiPathComponent in a variable and separate out the Ok(tp)
for clarity. No functional change.
|
| | |
| |
| |
| |
| | |
Call sites are shortly going to need the component on the error path,
so this function ought probably not to consume it.
|
| | |
| |
| |
| | |
Formatting churn at call sites.
|
| | |
| |
| |
| | |
The new function checks both directions. The (misspelled) macro didn't.
|
| | |
| |
| |
| | |
Now the only manual impl is TimePeriod, which is indeed special.
|
| | |
| |
| |
| | |
Mandatory horrible formatting.
|
| | | |
|
| | | |
|
| |/
|
|
|
|
| |
(This doesn't yet send any status events.)
Begins work on arti#1083.
|
| | |
|
| |
|
|
|
| |
Logging the current time and the start of the time period might help us
debug #1143 (we'll at least know how far off `now` is from `start`).
|
| | |
|
| | |
|
| | |
|
