Configuring OpenTripPlanner
Base Directory
On the OTP2 command line you must always specify a single directory after all the switches. This
tells OTP2 where to look for any configuration files. By default OTP will also scan this directory
for input files to build a graph (GTFS, OSM, elevation, and base street graphs) or the graph.obj
file to load when starting a server.
A typical OTP2 directory for a New York City graph might include the following:
otp-config.json
build-config.json
router-config.json
new-york-city-no-buildings.osm.pbf
nyc-elevation.tiff
long-island-rail-road.gtfs.zip
mta-new-york-city-transit.gtfs.zip
port-authority-of-new-york-new-jersey.gtfs.zip
graph.obj
You could have more than one of these directories if you are building separate graphs for separate
regions. Each one should contain one or more GTFS feeds, a PBF OpenStreetMap file, some JSON
configuration files, and any output files such as graph.obj
. For convenience, especially if you
work with only one graph at a time, you may want to place your OTP2 JAR file in this same directory.
Note that file types are detected through a case-insensitive combination of file extension and words
within the file name. GTFS file names must end in .zip
and contain the letters gtfs
, and OSM
files must end in .pbf
.
It is also possible to provide a list of input files in the configuration, which will override the default behavior of scanning the base directory for input files. Scanning is overridden independently for each file type, and can point to remote cloud storage with arbitrary URIs. See the storage section for further details.
Three Scopes of Configuration
OTP is configured via three configuration JSON files which are read from the directory specified on its command line. We try to provide sensible defaults for every option, so all three of these files are optional, as are all the options within each file. Each configuration file corresponds to options that are relevant at a particular phase of OTP usage.
Options and parameters that are taken into account during the graph building process will be "baked
into" the graph, and cannot be changed later in a running server. These are specified
in build-config.json
. Other details of OTP operation can be modified without rebuilding the graph.
These run-time configuration options are found in router-config.json
. Finally, otp-config.json
contains simple switches that enable or disable system-wide features.
Configuration types
The OTP configuration files use the JSON file format. OTP allows comments and unquoted field names
in the JSON configuration files to be more human-friendly. OTP supports all the basic JSON types:
nested objects {...}
, arrays []
, numbers 789.0
and boolean true
or false
. In addition to
these basic types some configuration parameters are parsed with some restrictions. In the
documentation below we will refer to the following types:
Type | Description | Examples |
---|---|---|
boolean |
This is the Boolean JSON type | true , false |
string |
This is the String JSON type. | "This is a string!" |
double |
A decimal floating point number. 64 bit. | 3.15 |
integer |
A decimal integer number. 32 bit. | 1 , -7 , 2100 |
long |
A decimal integer number. 64 bit. | -1234567890 |
enum |
A fixed set of string literals. | "RAIL" , "BUS" |
enum-map |
List of key/value pairs, where the key is a enum and the value can be any given type. | { "RAIL: 1.2, "BUS": 2.3 } |
enum-set |
List of enum string values | [ "RAIL", "TRAM" ] |
locale |
Language[\_country[\_variant]] . A Locale object represents a specific geographical, political, or cultural region. For more information see the Java Locale. |
"en_US" , "nn_NO" |
date |
Local date. The format is YYYY-MM-DD (ISO-8601). | "2020-09-21" |
date-or-period |
A local date, or a period relative to today. The local date has the format YYYY-MM-DD and the period has the format PnYnMnD or -PnYnMnD where n is a integer number. |
"P1Y" , "-P3M2D" , "P1D" |
duration |
A duration is a amount of time. The format is PnDTnHnMnS or nDnHnMnS where n is a integer number. The D (days), H (hours), M (minutes) and S (seconds) are not case sensitive. |
"3h" , "2m" , "1d5h2m3s" , "-P2dT-1s" |
regexp |
A regular expression pattern used to match a sting. | "$^" , "gtfs" , "\w{3})-.*\.xml" |
uri |
An URI path to a resource like a file or a URL. Relative URIs are resolved relative to the OTP base path. | "http://foo.bar/" , "file:///Users/jon/local/file" , "graph.obj" |
time-zone |
Time-Zone ID | "UTC" , "Europe/Paris" , "-05:00" |
feed-scoped-id |
FeedScopedId | "NO:1001" , "1:101" |
cost-linear-function |
A cost-linear-function used to calculate a cost from another cost or time/duration. Given a function of time: f(t) = a + b * t then a is the constant time part, b is the time-coefficient, and t is the variable. If a=0s and b=0.0 , then the cost is always 0 (zero). Examples: 0s + 2.5t , 10m + 0t and 1h5m59s + 9.9t The constant must be 0 or a positive number or duration. The unit is seconds unless specified using the duration format. A duration is automatically converted to a cost. The coefficient must be in range: [0.0, 100.0] |
|
time-penalty |
A time-penalty is used to add a penalty to the duration/arrival-time/departure-time for a path. It will be invisible to the end user, but used during the routing when comparing stop-arrival/paths. Given a function of time: f(t) = a + b * t then a is the constant time part, b is the time-coefficient, and t is the variable. If a=0s and b=0.0 , then the cost is always 0 (zero). Examples: 0s + 2.5t , 10m + 0 x and 1h5m59s + 9.9t The constant must be 0 or a positive number(seconds) or a duration. The coefficient must be in range: [0.0, 100.0] |
|
map |
List of key/value pairs, where the key is a string and the value can be any given type. | { "one": 1.2, "two": 2.3 } |
object |
Config object containing nested elements | "walk": { "speed": 1.3, "reluctance": 5 } |
array |
Config object containing an array/list of elements | "array": [ 1, 2, 3 ] |
System environment and project information substitution
OTP supports injecting system environment variables and project information parameters into the
configuration. A pattern like ${VAR_NAME}
in a configuration file is substituted with an
environment variable with name VAR_NAME
. The substitution is done BEFORE the JSON is parsed, so
both json keys and values are subject to substitution. This is useful if you want OTPs version number
to be part of the graph-file-name, or you want to inject credentials in a cloud based deployment.
{
"gsCredentials": "${GCS_SERVICE_CREDENTIALS}",
"graph": "file:///var/otp/graph-${otp.serialization.version.id}.obj"
}
In the example above the environment variable GCS_SERVICE_CREDENTIALS
on the local machine where
OTP is deployed is injected into the config. Also, the OTP serialization version id is injected.
The project information variables available are:
maven.version
maven.version.short
maven.version.major
maven.version.minor
maven.version.patch
maven.version.qualifier
git.branch
git.commit
git.commit.timestamp
graph.file.header
otp.serialization.version.id
Config version
All three configuration files have an optional configVersion
property. The property can be used to
version the configuration in a deployment pipeline. The configVersion
is not used by OTP in any
way, but is logged at startup and is available as part of the server-info data in the REST API.
The intended usage is to be able to check which version of the configuration the graph was build
with and which version the router uses. In an deployment with many OTP instances it can be useful to
ask an instance about the version, instead of tracking the deployment pipeline backwards to find the
version used. How you inject a version into the configuration file is up to you, but you can do it
in your build-pipeline, at deployment time or use system environment variable substitution.
OTP Serialization version id and Graph.obj file header
OTP has a OTP Serialization Version Id maintained in the pom.xml_ file. OTP stores the id in the serialized Graph.obj file header, allowing OTP to check for compatibility issues when loading the graph. The header info is available to configuration substitution:
${graph.file.header}
Will expand to:OpenTripPlannerGraph;0000007;
${otp.serialization.version.id}
Will expand to:7
The intended usage is to be able to have a graph build pipeline that "knows" the matching graph and OTP planner instance. For example, you may build new graphs for every OTP serialization version id in use by the planning OTP instances you have deployed and plan to deploy. This way you can roll forward and backward new OTP instances without worrying about building new graphs.
There are various ways to access this information. To get the Graph.obj
serialization version id
you can run the following bash command:
head -c 29 Graph.obj ==> OpenTripPlannerGraph;0000007;
(file header)head -c 28 Graph.obj | tail -c 7 ==> 0000007
(version id)
The Maven pom.xml, the META-INF/MANIFEST.MF, the OTP command line(--serializationVersionId
),
log start-up messages and all OTP APIs can be used to get the OTP Serialization Version Id.
Include file directive
It is possible to inject the contents of another file into a configuration file. This makes it
possible to keep parts of the configuration in separate files. To include the contents of a file,
use
${includeFile:FILE_NAME}
. The FILE_NAME
must be the name of a file in the configuration
directory. Relative paths are not supported.
To allow both files (the configuration file and the injected file) to be valid JSON files, a special
case is supported. If the include file directive is quoted, then the quotes are removed, if the
text inserted is valid JSON (starts with {
and ends with }
).
Variable substitution is performed on configuration file after the include file directive; Hence variable substitution is also performed on the text in the injected file.
Here is an example including variable substitution, assuming version 2.5.0 of OTP:
The result will look like this:
System-wide Configuration
Using the file otp-config.json
you can enable or disable different APIs and experimental
Sandbox Extensions. By default, all supported APIs are enabled and all
sandbox features are disabled. So for most OTP2 use cases it is not necessary to create this file.
Features that can be toggled in this file are generally only affect the routing phase of OTP2 usage,
but for consistency all such "feature flags", even those that would affect graph building, are
managed in this one file.
OTP Features
Here is a list of all features which can be toggled on/off and their default values.
Feature | Description | Enabled by default | Sandbox |
---|---|---|---|
APIBikeRental |
Enable the bike rental endpoint. | ✓️ | |
APIServerInfo |
Enable the server info endpoint. | ✓️ | |
APIUpdaterStatus |
Enable endpoint for graph updaters status. | ✓️ | |
ConsiderPatternsForDirectTransfers |
Enable limiting transfers so that there is only a single transfer to each pattern. | ✓️ | |
DebugUi |
Enable the debug GraphQL client and web UI and located at the root of the web server as well as the debug map tiles it uses. Be aware that the map tiles are not a stable API and can change without notice. Use the vector tiles feature if you want a stable map tiles API. | ✓️ | |
FloatingBike |
Enable floating bike routing. | ✓️ | |
GtfsGraphQlApi |
Enable the GTFS GraphQL API. | ✓️ | |
GtfsGraphQlApiRentalStationFuzzyMatching |
Does vehicleRentalStation query also allow ids that are not feed scoped. | ||
MinimumTransferTimeIsDefinitive |
If the minimum transfer time is a lower bound (default) or the definitive time for the transfer. Set this to true if you want to set a transfer time lower than what OTP derives from OSM data. |
||
OptimizeTransfers |
OTP will inspect all itineraries found and optimize where (which stops) the transfer will happen. Waiting time, priority and guaranteed transfers are taken into account. | ✓️ | |
ParallelRouting |
Enable performing parts of the trip planning in parallel. | ||
TransferConstraints |
Enforce transfers to happen according to the transfers.txt (GTFS) and Interchanges (NeTEx). Turning this off will increase the routing performance a little. | ✓️ | |
TransmodelGraphQlApi |
Enable the Transmodel (NeTEx) GraphQL API. | ✓️ | ✓️ |
ActuatorAPI |
Endpoint for actuators (service health status). | ✓️ | |
AsyncGraphQLFetchers |
Whether the @async annotation in the GraphQL schema should lead to the fetch being executed asynchronously. This allows batch or alias queries to run in parallel at the cost of consuming extra threads. | ||
Co2Emissions |
Enable the emissions sandbox module. | ✓️ | |
DataOverlay |
Enable usage of data overlay when calculating costs for the street network. | ✓️ | |
FaresV2 |
Enable import of GTFS-Fares v2 data. | ✓️ | |
FlexRouting |
Enable FLEX routing. | ✓️ | |
GoogleCloudStorage |
Enable Google Cloud Storage integration. | ✓️ | |
LegacyRestApi |
Enable legacy REST API. This API will be removed in the future. | ✓️ | ✓️ |
RealtimeResolver |
When routing with ignoreRealtimeUpdates=true, add an extra step which populates results with real-time data | ✓️ | |
ReportApi |
Enable the report API. | ✓️ | |
RestAPIPassInDefaultConfigAsJson |
Enable a default RouteRequest to be passed in as JSON on the REST API - FOR DEBUGGING ONLY! | ||
SandboxAPIGeocoder |
Enable the Geocoder API. | ✓️ | |
SandboxAPIMapboxVectorTilesApi |
Enable Mapbox vector tiles API. | ✓️ | |
SandboxAPIParkAndRideApi |
Enable park-and-ride endpoint. | ✓️ | |
SandboxAPITravelTime |
Enable the isochrone/travel time surface API. | ✓️ | |
TransferAnalyzer |
Analyze transfers during graph build. | ✓️ |
Example
JVM configuration
This section contains general recommendations for tuning the JVM in a production environment.
It focuses mainly on garbage collection configuration and memory settings.
See Garbage Collector Tuning for general information on garbage collection.
See Large Pages in Java and Transparent Huge Pages for general information on large memory pages.
OTP server
The OTP server processes concurrent routing requests in real time.
The main optimization goal for the OTP server is minimizing response time.
Garbage collector
- The G1 garbage collector (default since Java 9) offers a good compromise between low latency (i.e. low GC pause time) and GC overhead.
- If latency spikes are an issue, the ZGC garbage collector is an alternative. It produces in general more overhead than G1.
Memory settings
- Using Large Memory Pages can reduce pressure on the TLB cache and increase performance.
- It is in general not recommended to use large memory page in Transparent Huge Page mode (
-XX:+UseTransparentHugePages
) for latency-sensitive applications, since memory is allocated on-demand and this can induce latency spikes if the memory is fragmented.
Thus TLBFS mode (-XX:+UseHugeTLBFS
) should be the first choice. - If TLBFS mode is not an option, Transparent Huge Page mode (
-XX:+UseTransparentHugePages
) can be used instead, with additional provisions to mitigate the risk of latency spikes:
The physical memory can be committed upfront, at JVM startup time. This can be done by forcing a fixed heap size and pre-touching the memory.
Example:-Xms18g -Xmx18g -XX:+UseTransparentHugePages -XX:+AlwaysPreTouch
Graph Builder
The Graph Builder is the non-interactive mode used to build street graphs and transit graphs.
The main optimization goal for the Graph Builder is minimizing total build time.
Garbage collector
- In theory, the Parallel garbage collector offers the best throughput.
In practice, it can be challenging to optimize the Parallel GC to build both a street graph and a transit graph, the memory usage patterns being different. - The G1 garbage collector provides in general a good compromise.
Memory settings
- Using Large Memory Pages can reduce pressure on the TLB cache and increase performance.
- Since latency is not an issue, Large Memory Pages can be used indifferently in TLBFS mode (
-XX:+UseHugeTLBFS
) or Transparent Huge Page mode (-XX:+UseTransparentHugePages
)