This commit adds support for OpenCensus tracing
and metrics collection. This required support for
context.Context propogation throughout the cluster
codebase, and in particular, the ipfscluster component
interfaces.
The tracing propogates across RPC and HTTP boundaries.
The current default tracing backend is Jaeger.
The metrics currently exports the metrics exposed by
the opencensus http plugin as well as the pprof metrics
to a prometheus endpoint for scraping.
The current default metrics backend is Prometheus.
Metrics are currently exposed by default due to low
overhead, can be turned off if desired, whereas tracing
is off by default as it has a much higher performance
overhead, though the extent of the performance hit can be
adjusted with smaller sampling rates.
License: MIT
Signed-off-by: Adrian Lanzafame <adrianlanzafame92@gmail.com>
This changes the current strategy to extract headers from the IPFS daemon to
use them for hijacked endpoints in the proxy. The ipfs daemon is a bit of a
mess and what we were doing is not really reliable, specially when it comes to
setting CORS headers right (which we were not doing).
The new approach is:
* For every hijacked request, make an OPTIONS request to the same path, with
the given Origin, to the IPFS daemon and extract some CORS headers from
that. Use those in the hijacked response
* Avoid hijacking OPTIONS request, they should always go through so the IPFS
daemon controls all the CORS-preflight things as it wants.
* Similar to before, have a only-once-triggered request to extract other
interesting or custom headers from a fixed IPFS endpoint. This allows us to
have the proxy forward other custom headers and to catch
`Access-Control-Expose-Methods`. The difference is that the endpoint use for
this and the additional headers are configurable by the user (but with hidden
configuration options because this is quite exotic from regular usage).
Now the implementation:
* Replaced the standard Muxer with gorilla/mux (I have also taken the change
to update the gxed version to the latest tag). This gives us much better
matching control over routes and allows us to not handle OPTIONS requests.
* This allows also to remove the extractArgument code and have proper handlers
for the endpoints passing command arguments as the last segment of the URL. A
very simple handler that wraps the default ones can be used to extract the
argument from the url and put it in the query. Overall much cleaner this way.
* No longer capture interesting headers from any random proxied request. This
made things complicated with a wrapping handler. We will just trigger the one
request to do it when we need it.
* When preparing the headers for the hijacked responses:
* Trigger the OPTIONS request and figure out which CORS things we should set
* Set the additional headers (perhaps triggering a POST request to fetch them)
* Set our own headers.
* Moved all the headers stuff to a new headers.go file.
* Added configuration options (hidden by default) to:
* Customize the extract headers endpoint
* Customize what additional headers are extracted
* Use HTTPs when talking to the IPFS API
* I haven't tested this, but I did not want to have hardcoded 'http://' urls
around, as before.
* Added extra testing for this, and tested manually a lot comparing the
daemon original output with our hijacked endpoint outputs while looking
at the API traffic with ngrep and making sure the requets happen as expected.
Also tested with IPFS companion in FF and Chrome.
License: MIT
Signed-off-by: Hector Sanjuan <code@hector.link>
This commit makes the proxy extract useful fixed headers (like CORS) from
the IPFS daemon API responses and then apply them to the responses
from hijacked endpoints like /add or /repo/stat.
It does this by caching a list of headers from the first IPFS API
response which has them. If we have not performed any proxied request or
managed to obtain the headers we're interested in, this will try triggering a
request to "/api/v0/version" to obtain them first.
This should fix the issues with using Cluster proxy with IPFS Companion and
Chrome.
License: MIT
Signed-off-by: Hector Sanjuan <code@hector.link>
Removed unnecessary peername assignment
Modified tests according to the changes made to add peername to PinInfo
License: MIT
Signed-off-by: Kishan Mohanbhai Sagathiya <kishansagathiya@gmail.com>
No more hacks around /add. This uses the local adder when hijacking /add.
It supports the parameters and works pretty well with the ipfs CLI, showing
progress and everything.
License: MIT
Signed-off-by: Hector Sanjuan <code@hector.link>
tests cover local and sharded adds of files
ipfs mock and ipfs block put/get calls cleaned up
License: MIT
Signed-off-by: Wyatt Daviau <wdaviau@cs.stanford.edu>
deterministically sample bytes into files of different sizes
reorganized hash storage for easy access from all subpacks
import tests pass with new directory structure
License: MIT
Signed-off-by: Wyatt Daviau <wdaviau@cs.stanford.edu>
api client test
api test
one sharness test
refactoring of testingData for access from other packages
always wrap files added by cluster
remove unused flag 'progress'
changes to support hidden flag
License: MIT
Signed-off-by: Wyatt Daviau <wdaviau@cs.stanford.edu>
4 PinTypes specify how CID is pinned
Changes to Pin and Unpin to handle different PinTypes
Tests for different PinTypes
Migration for new state format using new Pin datastructures
Visibility of the PinTypes used internally limited by default
License: MIT
Signed-off-by: Wyatt Daviau <wdaviau@cs.stanford.edu>
This uses go-libp2p-kad-dht as routing provider for the Cluster Peers.
This means that:
* A cluster peer can discover other Cluster peers even if they are
not in their peerstore file.
* We remove a bunch of code sending and receiving peers multiaddresses
when a new peer was added to the Cluster.
* PeerAdd now takes an ID and not a multiaddress. We do not need to
ask the new peer which is our external multiaddress nor broadcast
the new multiaddress to everyone. This will fix problems when bootstrapping
a new peer to the Cluster while not all the other peers are online.
* Adding a new peer does not mean to open connections to all peers
anymore. The number of connections will be made according to the DHT
parameters (this is good to have for future work)
The that detecting a peer addition in the watchPeers() function does
no longer mean that we have connected to it or that we know its
multiaddresses. Therefore it's no point to save the peerstore in these
events anymore.
Here a question opens, should we save the peerstore at all, and should we
save multiaddresses only for cluster peers, or for everyone known?
Currently, the peerstore is only updated on clean shutdown,
and it is updated with all the multiaddresses known, and not limited to
peer IDs in the cluster, (because, why not).
License: MIT
Signed-off-by: Hector Sanjuan <code@hector.link>
This commit attempts to fix race issues in the maptracker since the
introduction of the OperationTracker.
There were two main problems:
* Duplicity tracking the state both in the state map and the opTracker
* Non atomiciy of operations with different threads being able to affect
other threads operations.
A test performing random Track/Untracks on the same Cid quickly showed
that items would sometimes stay as pin_queued or pin_unqueued. That happened
because operations could be cancelled under the hood by a different request,
while leaving the map status untouched.
It was not simply to deal with this issues without a refactoring.
First, the state map has been removed, and the operation tracker now provides
status information for any Cid. This implies that the tracker keeps all
operations and operations have a `PhaseDone`. There's also a
new `OperationRemote` type.
Secondly, operations are only created in the tracker and can only be removed
by their creators (they can be overwritten by other operations though).
Operations cannot be accessed directly and modifications are limited to setting
Error for PhaseDone operations.
After created, *Operations are queued in the pinWorker queues which handle any
status updates. This means, that, even when an operation has been removed from
the tracker, status updates will not interfere with any other newer operations.
In the maptracker, only the Unpin worker Cleans operations once processed. A
sucessful unpin is the only way that a delete() happens in the tracker map.
Otherwise, operations stay there until a newer operation for the Cid arrives
and 1) cancels the existing one 2) takes its place. The tracker refuses to
create a new operation if a similar "ongoing" operation of the same type
exists.
The final change is that Recover and RecoverAll() are not async and play by the
same rules as Track() and Untrack(), queueing the items to be recovered.
Note: for stateless pintracker, the tracker will need to Clean() operation
of type OperationPin as well, and complement the Status reported
by the tracker with those coming from IPFS.
License: MIT
Signed-off-by: Hector Sanjuan <code@hector.link>
