In principle a connection can close while it's in progress with
starting, and then it's undefined if we wait for goroutines to exit etc.
With this change, we will wait for start to complete before starting to
stop everything.
This adds the ability to have multiple concurrent connections to a single device. This is primarily useful when the network has multiple physical links for aggregated bandwidth. A single connection will never see a higher rate than a single link can give, but multiple connections are load-balanced over multiple links.
It is also incidentally useful for older multi-core CPUs, where bandwidth could be limited by the TLS performance of a single CPU core -- using multiple connections achieves concurrency in the required crypto calculations...
Co-authored-by: Simon Frei <freisim93@gmail.com>
Co-authored-by: tomasz1986 <twilczynski@naver.com>
Co-authored-by: bt90 <btom1990@googlemail.com>
This adds a cache to the expensive key generation operations. It's fixes
size LRU/MRU stuff to keep memory usage bounded under absurd conditions.
Also closes#8600.
* lib/connections: Cache isLAN decision for later external access.
The check whether a remote device's address is on a local network
currently happens when handling the Hello message, to configure the
limiters. Save the result to the ConnectionInfo and pass it out as
part of the model's ConnectionInfo struct in ConnectionStats().
* gui: Use provided connection attribute to distinguish LAN / WAN.
Replace the dumb IP address check which didn't catch common cases and
actually could contradict what the backend decided. That could have
been confusing if the GUI says WAN, but the limiter is not actually
applied because the backend thinks it's a LAN.
Add strings for QUIC and relay connections to also differentiate
between LAN and WAN.
* gui: Redefine reception level icons for all connection types.
Move the mapping to the JS code, as it is much easier to handle
multiple switch cases by fall-through there.
QUIC is regarded no less than TCP anymore. LAN and WAN make the
difference between levels 4 / 3 and 2 / 1:
{TCP,QUIC} LAN --> {TCP,QUIC} WAN --> Relay LAN --> Relay WAN -->
Disconnected.
This replaces old style errors.Wrap with modern fmt.Errorf and removes
the (direct) dependency on github.com/pkg/errors. A couple of cases are
adjusted by hand as previously errors.Wrap(nil, ...) would return nil,
which is not what fmt.Errorf does.
Having a separate mutex for the three or four instructions needed to
fetch and increment nextID means the overhead exceeds the cost of this
operation. nextID is now handled inside the critical section for
awaiting instead, while the more expensive channel creation has been
moved outside it.
This is mostly a simplification, though it may have minor performance
benefits in some situations. The single-threaded sender benchmark shows
no significant difference:
name old speed new speed delta
RequestsRawTCP-8 55.3MB/s ± 7% 56.6MB/s ± 6% ~ (p=0.190 n=10+10)
RequestsTLSoTCP-8 20.5MB/s ±20% 20.8MB/s ± 8% ~ (p=0.604 n=10+9)
* lib/protocol: Require at least 3.125% savings from compression
The new lz4 library doesn't need its output buffer to be the maximum
size, unlike the old one (which would allocate if it weren't). It can
take a buffer that is of a smaller size and will report if compressed
data can fit inside the buffer (with a small chance of reporting a false
negative). Use that property to our advantage by requiring compressed
data to be at most n-n/32 = .96875*n bytes long for n input bytes.
* lib/protocol: Remove unused receivers
To make DeepSource happy.
* lib/protocol: Micro-optimize lz4Compress
Only write the length if compression was successful. This is a memory
write, so the compiler can't reorder it.
Only check the return value of lz4.CompressBlock. Length-zero inputs
are always expanded by LZ4 compression (the library documents this),
so the check on len(src) isn't needed.
* lib/model: Remove bogus fields from connections API endpoint.
Switch the returned data type for the /rest/system/connections element
"total" to use only the Statistics struct. The other fields of the
ConnectionInfo struct are not populated and misleading.
* Lowercase JSON field names.
* lib/model: Get rid of ConnectionInfo.MarshalJSON().
It was missing the StartedAt field from the embedded Statistics
struct. Just lowercasing the JSON attribute names can be done just as
easily with annotations.
* lib/model: Remove bogus startedAt field from totals.
Instead of using the Statistics type with one field empty, just switch
to a free-form map with the three needed fields.
This truncates times meant for API consumption to second precision,
where fractions won't typically matter or add any value. Exception to
this is timestamps on logs and events, and of course I'm not touching
things like file metadata.
I'm not 100% certain this is an exhaustive change, but it's the things I
found by grepping and following the breadcrumbs from lib/api...
I also considered general-but-ugly solutions, like having the API
serializer itself do reflection magic or even regexps on returned
objects, but decided against it because aurgh...
This adds a statistic to track the last connection duration per device.
It isn't used for much in this PR, but it's available for #7223 to use
in deciding how to order device connection attempts (deprioritizing
devices that just dropped our connection the last time).
This adds error returns to model methods called by the protocol layer.
Returning an error will cause the connection to be torn down as the
message couldn't be handled. Using this to signal that a folder isn't
currently available will then cause a reconnection a few moments later,
when it'll hopefully work better.
Tested manually by running with STRECHECKDBEVERY=0 on a nontrivially
sized setup. This panics reliably before this patch, but just causes a
disconnect/reconnect now.
We incorrectly gave a too small buffer to lz4.Compress, causing it to
allocate in some cases (when the data actually becomes larger when
compressed). This then panicked when passed to the buffer pool.
This ensures a buffer that is large enough, and adds tripwires closer to
the source in case this ever pops up again. There is a test that
exercises the issue.
* lib/protocol: Wait for reader/writer loops on close (fixes#4170)
* waitgroup
* lib/model: Don't hold lock while closing connection
* fix comments
* review (lock once, func argument) and naming
* lib/model: Send cluster config before releasing pmut
* reshuffle
* add model.connReady to track cluster-config status
* Corrected comments/strings
* do it in protocol
This constructs the map of hashes of zero blocks from constants instead
of calculating it at startup time. A new test verifies that the map is
correct.
