A request to osrm-routed can be assigned to a thread which
is currently busy processing another request, even when there
are other threads/cores available. This unnecessarily delays
the response, and can make requests appear to hang when
awaiting CPU intensive requests to finish.
The issue looks like a bug in Boost.Asio multithreaded
networking stack.
osrm-routed server implementation is
heavily influenced by the HTTP server 3 example in the
Boost.Asio docs. By upgrading to Boost 1.70 and updating the
server connections to match the example provided in the 1.70
release, the problem is resolved.
The diff of the changes to the Boost.Asio stack are
vast, so it's difficult to identify the exact cause. However
the implementation change is to push the strand of execution
into the socket (and timer) objects, which suggests it could
fix the type of threading issue we are observing.
Each MLD cell has source and destination nodes.
MLD is keeping a |source| x |destination| sized table
for various metrics (distances, durations, etc) from each
source to all destinations in a cell.
It stores all of the values for a metric in one large array, with
an offset for each cell to find its values. The offset is currently
limited to 32 bit values, which overflows on very large graphs
(e.g. Planet OSM).
We fix this by changing the offsets to be uint64_t types.
The generation of level masks for compactly storing partition cells
supports sizes that can be stored in 64 bits.
The current implementation fails if the total bit sum is 64 bits
exactly. A bit shift mechanism is used that is undefined when the
shift size is equal to the bit size of the underlying type. This
generates an incorrect mask value.
We fix this by adding a special case for a 64 bit offset. Given this
code is called at most |level| times, there will be no effect on
performance. We also update the assertions to reflect 64 bit masks
are now supported.
osrm-routed does not immediately clean up a keep-alive connection
when the client closes it. Instead it waits for five seconds
of inactivity before removing.
Given a setup with low file limits and clients opening and
closing a lot of keep-alive connections, it's possible for
osrm-routed to run out of file descriptors whilst it waits for
the clean-up to trigger.
Furthermore, this causes the connection acceptor loop to exit.
Even after the old connections are cleaned up, new ones
will not be created. Any new requests will block until the
server is restarted.
This commit improves the situation by:
- Immediately closing connections on error. This includes EOF errors
indicating that the client has closed the connection. This releases
resources early (including the open file) and doesn't wait for the
timer.
- Log when the acceptor loop exits. Whilst this means the behaviour
can still occur for reasons other than too many open files,
we will at least have visibility of the cause and can investigate further.
OSM node 2^33 was created in early April 2021. This and all
subsequently created IDs will be overflowing OSRM node storage
which only support 33 bit IDs.
Bump the number of bits to 34 to double node ID capacity. This
is a breaking change to the data format as it alters the layout
of .osrm.nbg_nodes.
Removes the breaking libosrm API change by adding the old interface to
the new. This does not introduce any new breaks.
The downside of this is that it allows for multiple ways to
return JSON responses.
When using non-default constructors for the API parameter classes,
vector arguments like coordinates and hints are copied at least once
(twice when passed as lvalue arguments).
Enable perfect forwarding of BaseParameter arguments and
pass-by-value in the constructor that uses the argument. This
ensures we copy at most once (zero for rvalue arguments).
Currently OSRM only supports turn restrictions with a single via-node or one
via-way. OSM allows for multiple via-ways to represent longer and more
complex restrictions.
This PR extends the use of duplicate nodes for representng via-way turn
restrictions to also support multi via-way restrictions. Effectively, this
increases the edge-based graph size by the number of edges in multi via-way
restrictions. However, given the low number of these restrictions it
has little effect on total graph size.
In addition, we add a new step in the extraction phase that constructs
a restriction graph to support more complex relationships between restrictions,
such as nested restrictions and overlapping restrictions.
* Use Nan:: getters and setters for node 12/14 compatibility
* Drop support for publishing node 8 bindings, add publishing support for node 12, 14
Co-authored-by: Daniel Patterson <danpat@danpat.net>
As part of graph contraction, node renumbering leads to
in-place permuting of graph state, including boolean vector elements.
std::vector<bool> returns proxy objects when referencing individual
bits. To correctly swap bool elements using MSVC, we need to explicitly
apply std::vector<bool>::swap.
Making this change fixes osrm-contract on Windows.
We also correct failing tests and other undefined behaviours
(mainly iterator access outside boundaries) highlighted by MSVC.
For very large graphs, generation of MLD level masks fail on Windows
due to bit shift overflow of unsigned long values.
Correct by using unsigned long long literals, which are 64 bit on
all major systems.
In cases where we are unable to find a phantom node for an input
coordinate, we return an error indicating which coordinate failed.
This would always refer to the coordinate with index equal to the
number of valid phantom nodes found.
We fix this by instead returning the first index for which a
phantom node could not be found.
