Previously, when the larva got to processing enqueued events, it was
doing so without loading state into the adult beforehand, resulting in
incorrect processing of events.
Here, we make the larva call +molt more eagerly, ensuring that the adult
always has its state available when we use it.
Yes, there is a global timer for closing flows, but all that does is
enqueue a cork message. +on-stir needs to set _pump_ timers for all
flows that might still have messages to send, which includes closing
flows.
When ames notifies us that our subscription has been kicked, we enqueue
a cork to clean up the flow. Unlike the %leave case, however, we were
not registering the cork in the queue of outstanding comms. We would
eventually get an ack, but not know what for, and erroneously inject
%poke-acks and %watch-acks.
Here we simply add a %cork entry to the queue before sending it.
This is sufficient to bring the normal (non-prerelease-bugged) cases
into the new world.
For the prerelease ships that ran a buggier version of the new gall
subscription logic, we note that the conditional may trigger for the
nonce=1 case where it had already triggered for their
(shouldn't-be-possible) nonce=0 case. This results in a %leave on a wire
that wasn't in use. This no-ops on the publisher side though, and the
flow gets corked right away, so this is considered harmless.
In response to clog notification from remote ames, we were sending a
%cork to clean up the flow. However, the wire we were using had the /sys
prefix already stripped off. Here, we put it back in.
Start by killing subscription nonce 0, then work our way up instead of
down. We enhance the printf with a "total nonces" indicator so we can
still easily see the progress being made.
Previous +ap-doff kicked the agent repeatedly. We needed to kick
it only once. Now publisher agents clear their incoming subscription
state without the subscriber making lots of new subscriptions because
of repeated kicking.
+on-plea gets called in two very different ways:
1) handling request from local vane to send %plea to peer
2) handling %cork request from another ship, which our local ames has %pass'ed
to ourselves
In the second case, we shouldn't print misleadingly, or bind a duct in the ossuary.
+ap-nuke was not including the nonce, but should.
+ap-handle-peers was potentially including a zero nonce.
(The latter shouldn't have been possible, but there's a bug in +load
where sub-nonce.yoke gets initialized as 0 instead of 1.)
Gall tells ames to %cork flows for subscriptions it has closed.
Receiving a kick also closes a subscription, but gall wasn't issuing a
%cork in that case. We correct that here.
Inlines +mo-handle-ames-response's logic at its only callsite.
seems that this structure has been unused since
e75ab631a4 and confuses
newbies trying to figure out exactly what the commit
structure is (which is how I came across this)
Without this, a ship would send a cork on a max of one flow per
recork timer, which could take years to clear for some ships.
This starts a hot loop of trying the next cork once one gets
positively acked.
The previous recork timer queued up %cork messages without sending them.
It also relied on making sure pump timers didn't get set for recork bones.
This was fragile.
The new design enqueues up to one new %cork message per ship during each
recork timer, based on the state of the flow. If the flow is closing but
there are no outstanding messages in it, then it needs to be recorked.
Flows will be recorked in ascending numerical order by bone.
The condition got butchered during refactor: instead of avoiding the creation
of pump timers during recork wake, it was setting them _exclusively_ during
recork wake.
Problem:
by-channel has its own copy of server-state from line 2182. discard-channel returns an altered state, with one channel removed from the state of by-channel.
but the state of by-channel isn't changing with each iteration, so |trim is only removing one channel per invocation.
Solution:
update by-channel on each iteration.
This is a temporary fix, and first part of the gall-request-queue-fix
release in two stages. This gives a publisher ship the ability to
understand a %cork and handle it properly, but no subscriber will
be sending %corks at this stage when leaving a subscription.
We still add a nonce to all subscription wires but it doesn't
increment it when resubscribing, allowing flows to be reused.
Tested locally with toy pub/sub agents and Group join/leaving
Previously, the initial Azimuth snapshot was stored in Clay and shipped
in the pill. This causes several problems:
- It bloats the pill
- Updating the snapshot added large blobs to Clay's state. Even now
that tombstoning is possible, you don't want to have to do that
regularly.
- As a result, the snapshot was never updated.
- Even if you did tombstone those files, it could only be updated as
often as the pill
- And those updates would be sent over the network to people who didn't
need them
This moves the snapshot out of the pill and refactors Azimuth's
initialization process. On boot, when app/azimuth starts up, it first
downloads a snapshot from bootstrap.urbit.org and uses that to
initialize its state. As before, updates after this initial snapshot
come from an Ethereum node directly and are verified locally.
Relevant commands are:
- `-azimuth-snap-state %filename` creates a snapshot file
- `-azimuth-load "url"` downloads and inits from a snapshot, with url
defaulting to https://bootstrap.urbit.org/mainnet.azimuth-snapshot
- `:azimuth &azimuth-poke-data %load snap-state` takes a snap-state any
way you have it
Note the snapshot is downloaded from the same place as the pill, so this
doesn't introduce additional trust beyond what was already required.
When remote scry is released, we should consider allowing downloading
the snapshot in that way.
Includes patched versions of ames' and clay's +load arms.
In clay, we do a dumb ;; hack to get the state to adapt properly. This
shouldn't be needed ($case had an extra... case added to it, old ones
should still nest), and so we should revisit the logic there to make it
cleaner/better before release.
Because the publisher will send the cork plea back to the subscriber on
the next bone, we are not able to know the bone for the original cork.
To handle it, we add the cork bone to the plea path
still wip: it keeps resending the cork plea faster than its ~h1 timer
Removes the !: at the top of gall, so that it no longer gets included in traces about agent builds or crashes.
We also refine intentional crashes with ~_s, so that we still see a crash reason even if we don't get a full trace.
Lastly, flops the trace for +on-load crashes, which were getting printed bottom-first.
Regularized arm names to +etch-* and +sift-* to match Vere. Renamed
$packet to $shot. Used $meow, $purr, and $keen to match Vere's naming.
Reorganized packet decoding arms and moved some to Lull for later use in
Aqua.
Stale lanes may cause forwarding loops. Imagine the following:
1) Planet A is live. Galaxy B, its indirect sponsor, learns of its route.
2) A goes offline. Another ship, C, is started in its place, at the same route.
3) B receives a packet for A, forwards it to the known route.
4) C received the packet, forwards it to B.
5) Repeat from 3.
Here, we update the forward lane(s) scry used by the runtime to not produce a
peer's lane if they haven't communicated with us in the last hour. Everyone's
supposed to ping their sponsorship chain every 30 seconds. If those aren't
going through, you shouldn't expect to be reachable anyway.
We may or may not want to update +send-blob to match.
Previously, if the pointer for a syntax error pointed to the end of the file
(and the file ended in a newline) the code snippet rendering would try to
display a line _beyond_ the end of the file, causing a crash.
Here, we detect that case, and display `<<end of file>>` instead.
Instead of having a separate "busy" flag and pending scry request state,
we now have a single "busy" unit that, if there is a pending request,
contains details about the pending request. In the ames case, this is
simply %ames. But in the scry case, it contains all the details we need
to cancel the request, timer, or both when needed.
Additionally, we now make sure to always cancel the scry timeout timer
whenever we get a scry response.
Previously, when the scry timeout fired and we retried the request with
ames, we would always start over from the very first request (ie,
fetching the ""manifest"" containing the files we need to download).
This behavior is not correct in the case where we had already received
that initial response, and were now fetching individual files. So here,
if that's the case, we simply call into +work after marking this peer as
having known-broken scry, which will resume work as appropriate.
(The actual bug here that broke the fallback behavior entirely was the
busy flag not getting unset. We now clean it up properly.)
We also move setting of the sad timer into +retry-with-ames, instead of
doing it at each individual callsite. (In fact, one of the callsites was
missing this behavior.)
This accounts for a possible race condition where ames expects a
response, but regresses into the larval state. Upon receiving the
$sign on +take, we would remain stuck as a larva. Now we check
that we have enough information to re-evolve and then start a
/larval timer to begin draining the queue.
When +ke-abet-gone gets called, we're going to remove this keen
entirely. +ke-set-wake does a whole song-and-dance and may even set a
new timer. So instead, we simply call +ke-rest, if we have a known
outstanding timer.
Previously we stored the nonce in $boat, which changed the $bowl of each
agent. This compiles and all agents reload, but more testing is needed.
It also renames inbound/outbound watches to $bitt/$boat.
Similar to how %x on the empty desk was already treated specially. We
continue supporting %d on non-empty desks, but do add a print marking it
as deprecated.
The previous commit already started treating the 0v0 commit as
equivalent to the 0 aeon (that is, never-real root), so we should give
the 0v0 commit for all 0 aeons.
Instead of scrying for revisions of files at paths in desks, we now scry
for data corresponding to a given lobe. This removes concerns of aeons,
paths, and other such frivolities, and lets us ask for the specific data
we need regardless of where it may live.
How we're supposed to answer permissioning questions around data this
way remains to be seen.
Previously we were dropping events that used old
wires that lacked a rift in them. This seems a
bad behavior because we don't want to destroy a
flow that has not been processed by both ends.
Note: pending a fix to test-old-ames-wire
This converts the blob store from having deltas, directs, and
tombstones, to just having direct pages. This simplifies a lot of code,
since we don't have to constantly ensure that deltas always have their
parent available.
This removes the hardcoded text diff logic from clay, which was
previously required for bootstrapping.
Over the wire, we handle both old and new requests and responses
transparently, so communication is normal in both directions across
ships which do or do not have this change.
We had trie operations independently implemented in +de in arvo,
+an:cloy in zuse, +zu in clay, lib/trie, and app/spider. This unifies
them all into +de in arvo, aggregating the used operations.
This fixes a space leak where the entire ford/clay core would be
included in the ford cache. Heavily reduces memory usage by clay,
reducing total usage from 100-200MB to around 10MB.
Instead of reporting a single memory size for built files, marks and
conversions, we now report memory size per path, mark name and mark
pair, respectively.
instead make it a rolling 128-bit integer. 128 bits is the same size as
the +sham space, so is one natural choice for "big enough to never have
to think about this." 64, 32, 16, even 8 bits would probably be fine.
Threads should eventually take and produce $cage instead of $vase. Since
%khan is likely to be used by third parties, we write to the eventual
intended API. We ignore the mark on the input $cage (it is safe to
always specify %noun), and we always use %noun as the output mark.
%fyrd now makes more sense. It was previously discarding the type of the
output %arow and re-encoding the raw noun as a vase of the output mark;
it is now performing mark conversion from the mark of the output $cage
to the originally requested output mark.
Use the correct wires when cancelling the scry request and/or its
timers.
Note that we may produce more %rests and %yawns than strictly necessary,
but these no-op cleanly in those cases.
tid was accidentally getting set to the name of the output mark. As we
don't currently support cancelling threads, there is no reason to
maintain the originally-intended correspondence between tid and conn
request-id.
Take the opportunity to clean up indentation somewhat.
This can occur after upgrade, or when receiving a %delta blob. We do not
know the path or commit timestamp of a file that resolves to the blob,
so we simply fall back to old ames-style blob-by-lobe fetching.
In +work, we simply want to fetch file data, regardless of what the
overarching request is. (In fact, %sing would likely never hit this path
anyway.)
In order to be able to make the scry request, we track the timestamp of
the commit that contains a blob we're missing, and scry for that %da
revision.
There's two edge cases where we cannot immediately know the timestamp
that are currently assumed broken. Fix soon.
%v requests are already handled specially for the ames case. We should
continue to respect that, only doing those kinds of requests over ames.
Also cancels the scry timer after receiving a response, instead of not
doing that.
Ancient version upgraders beware! If you're coming from a home-based
arvo, you must first upgrade to the version prior to this, or else
you'll be in trouble!
When clay wants to download blobs from a foreign ship, it attempts doing
this using the new remote scry protocol.
If it doesn't receive a response within ~m1, it falls back to using the
old ames-based syncing instead. We remember this "prefer ames" state for
the specific ship for an hour, after which we'll begin trying the scry
flow again.
Compiles, but untested. Some TODOs and REVIEWs remaining herein.
We intentionally leave the dist-upgraded flag in state to avoid
cluttering the diff here. The next commit will remove it.
Also strips out `$` from khan top-level comment.
There are arguments for keeping $crag in lull, and on the other side for
moving $cast to arvo. This seemed like the most reasonable approach.
%fyrd is now implemented in terms of %fard, and likewise %avow in terms
of %arow. State is tracked via wire rather than in a global map.
Unit tests adjusted to match.
These take and produce vases, and assign random tids (rather than
deducing them from the input duct.)
Since %fard does not require mark conversion, we make the mark/beak on
$thread-state optional (and use this to decide whether to send %avow or
%arow.) Provide a state adapter since it's possible that people have
been experimenting with this vane.
This makes the negative case of %avow/%arow kind of clunky, since there
is no content difference, but the following does not seem possible
within the Hoon type system:
=/ gif
?~ p.tad
%arow %avow
[hen %give gif %| p.cag tang]~
- use desk parameter instead of %base everywhere
- formatting clean up
- make |story-remove take a case instead of an aeon
- make desk param optional for story-set and story-log
Since we cannot scry the key out of jael during load, and the
alternative is defering this logic through the larval core in some way,
we simply include the type for the old acru interface, letting us reuse
the old core to initialize the new one.
Instead of exporting keys so that caller can do this themselves, we
expose arms for signing and authenticating that produce and operate on
just the signature, without mangling it into the message.
- only store metadata in the persistent map. just enough to support
(eventual) thread cancellation and output mark lookup.
- try to delete thread state at other failure points not covered by
%kick.
- reflect back the passed output mark rather than form.dais. not sure
about this one yet.
Resolves a good number of conflicts. Most notably, re-propagates removal
of gall's %onto, confirms new /app/herm behavior, coerces hood/drum
state adapters back into place, and updates webterm to use the latest
api.
By moving this behavior for packets with ames-style headers into lull,
other vanes that want to do protocols in similar style (like the upcoming
%fine vane) can reuse the logic here.
Note that we parameterize the main en- and decoding functions, so that
we may pass in the "is ames" flag and protocol versions as appropriate
to the context.
This change greatly improves the ergonomics of working with channel JSON
in statically typed languages, as the polymorphism is moved out of the
actual diff and into the event framing.
+wake had accumulated several layers of abstractions which were later
rendered unnecessary. This removes those abstractions and should have
no semantic effect.
This adds support for tombstoned files to clay. It does not include any
way to actually tombstone them; that is left for later.
This allows tombstoning at the level of a file. Precisely, this expands
+blob:clay by adding a %dead case:
+$ blob :: fs blob
$% [%delta p=lobe q=[p=mark q=lobe] r=page] :: delta on q
[%direct p=lobe q=page] :: immediate
[%dead p=lobe ~] :: tombstone
== ::
Thus, we maintain the invariant that every lobe corresponds to a blob,
but now a blob may be an explicit tombstone.
Details:
- This has not been tested at all, except that it compiles and boots.
- This does not have a state adapter from master. The only state change
is the definition of +cach.
- Additionally, out-of-date ships may unexpectedly receive a %dead blob
from a foreign clay which would interfere with their ability to download
that desk. No code changes necessary, but sponsors should avoid
tombstoning files in %base for a while so their children can get the
update.
- A merge will only fail if the tombstoned file conflicts with another
change. Note that as written, merging from a past desk *can* bring a
tombstoned file to the head of a desk. Possibly this shouldn't be
allowed.
This also includes a couple refactors that were made possible by ford
fusion (since everything is synchronous now) but never got done. In
both cases we get to remove a monad, which simplifies the code
considerably.
- refactor +merge's error handling to use !!/mule instead of threading
through errors
- refactor all +read-* functions and related parts of +try-fill-sub to
eagerly convert lobes to cages.
We also add support reading %a/b/c/e/f/r/x from past and foreign desks,
when possible. Apologies that all of these are in one commit, it was
all a single chunk of work.
This is a draft until we have a way to tombstone. I suspect we'll want
to have a mechanism of keeping track of gc roots and trace to remove,
but this PR doesn't suggest any particular strategy.
Jael needs to be reconfigured to listen to the new aagent for azimuth
events, and the old app needs to be shut down. We do this in
/app/azimuth's +on-init.
Additionally, we make sure that jael doesn't crash when it (as expected)
loses its subscription to the old agent.
When you loaded an app with an error, then fixed the error, it would
create the main gall %mult subscription at a time in the past. Then,
clay would never fill the subscription since it couldn't get the old %a
entries for the apps.
This fixes the issue in two ways: first, don't subscribe in the past.
Second, if clay can't get the old versions, just fire the subscription
anyway.
Previously, if trying to bind to an endpoint that was already bound to,
eyre would reject it. This doesn't play very nicely in a softdist world
where uninstalled apps might not get a chance to clean up, and apps
might re-bind simply for being re-installed.
Here we change eyre to overwrite an existing binding if it conflicts
with the new one to be added.
