[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [geomesa-users] Bug with geomesa.index.dtg and not allowing dates before the Epoch
|
It does look like we re-add the default date even if it is deleted from
the metadata :/ I'll open a ticket for that as well.
For now, I think your best bet is to just go with the dummy date field.
Thanks,
Emilio
On 04/24/2017 09:30 AM, Emilio Lahr-Vivaz wrote:
Hi David,
I opened a ticket to support dates before the epoch here:
https://geomesa.atlassian.net/browse/GEOMESA-1785
In general we assume that indicating a date field to index shouldn't
be mandatory, as users are likely to miss that step and suffer
degraded performance. It does seem useful to have some way to indicate
explicitly to not index any date though.
I believe that modifying the 'attributes' entry in the metadata table
should work though - try removing the entire 'geomesa.index.dtg'
entry, and make sure to bounce your application as we cache feature
types. I'll test it out to verify.
Thanks,
Emilio
On 04/21/2017 08:07 PM, David Boyd wrote:
I would like to report this as a bug. I am using 1.3.1 of Geomesa.
Here is the situation:
1. I have Date fields in my data with times before the start of the
Epoch.
2. I do not necessarily need them indexed within the Z3 indexes.
I set the hint to turn off the Z3 indexes - That works.
If I don't set geomesa.index.dtg or set it to null, GEOMESA picks a
field automatically to index.
The problem is it picks my fields with Dates before the Epoch.
So when I go to store the record it errors out because of a date out
of range (basically any date before the Epoch
will have a negative number ) and the range check fails.
I think at least one of the following fixes is needed:
1. Accept dates for indexing from before the Epoch (preferred, that
would let historical data to be time indexed).
2. Accept null for geomesa.index.dtg and don't use a date (in the
code if the that attribute is null it will use 0 for the
time. Except that the geomesa.index.dtg gets overwritten before that
when the schema is created).
3. If the z3 indexes are disabled stop doing the range check on the
date.
I have tried changing the <FeatureName>~Attributes record in the
metadata table to set the geomesa.index.dtg to null but
that seemed to be ignored.
I have hacked around it by adding a dummy date field to my date (not
ideal by any means).
I am not a scala developer, and if I had time on my project I would
attempt #2 or #3 above. But #1 would seem to be the right
fix.
Or am I missing something. This issue has cost me at least two full
days.
On 4/21/17 5:24 PM, David Boyd wrote:
Emilio:
So I found part of the problem.
I missspelled geomesa.indexes.enable (had geomeas).
So now at least my metadata attributes do not have the Z3 entry:
ActorRecordset~attributes : []
objectKey:String,entityName:String,entitySource:String,entityTitle:String,recordKey:String:cardinality=high:index=full,Name:String:cardinality=high:index=full,Type:String:cardinality=high:index=full,NameMetaphone:String:cardinality=high:index=full,Country:String:cardinality=high:index=full,AffiliationTo:String:cardinality=high:index=full,AffiliationStart:Date:cardinality=high:index=full,AffiliationEnd:Date:cardinality=high:index=full,Aliases:String:cardinality=high:index=full,GeoCountryCode:String:cardinality=high:index=full,*GeoLocation:Point;geomesa.index.dtg='AffiliationStart',geomesa.table.sharing='true',geomesa.indices='z2:3:3,records:2:3,attr:4:3',geomesa.table.sharing.prefix='\\u0002'
These are the tables I now have in accumulo:
CoalesceSearch
CoalesceSearch_attr_v4
CoalesceSearch_queries
CoalesceSearch_records_v2
CoalesceSearch_stats
CoalesceSearch_z2_v3
But I am still getting the validation error. It is picking the
AffiliationStart field as the first one for the default date index.
But that field is one that has the dates before the Epoch.
This is causing me a some real issues. I don't want to have to
clutter up my data with a dummy date field. I already have to create
dummy geometry fields for records with no location information.
If the Z3 indices are turned off why am I still getting validation
errors for the date? It should never be used.
On 4/21/17 4:48 PM, Emilio Lahr-Vivaz wrote:
Ok, since by default feature types will share a table, you can
expect to still see the _z3 table. I think somehow the user data is
not getting set right
before the call to createSchema. If you look at the 'atttributes'
row, you should see something like:
...geomesa.indices='z2:3:3,records:2:3,attr:4:3'...
(it shouldn't include the z3 entry).
You can try remote debugging to figure out what's wrong, this is
the line that should be handling it:
https://github.com/locationtech/geomesa/blob/master/geomesa-accumulo/geomesa-accumulo-datastore/src/main/scala/org/locationtech/geomesa/accumulo/data/AccumuloDataStore.scala#L185
As a work-around, you can edit the 'attributes' row through the
accumulo shell to remove the z3 reference - that will cause it to
stop writing and reading from z3.
Thanks,
Emilio
On 04/21/2017 04:34 PM, David Boyd wrote:
Emilio:
There are three feature types defined.
ActorRecordset~attributes : []
objectKey:String,entityName:String,entitySource:String,entityTitle:String,recordKey:String:cardinality=high:index=full,Name:String:cardinality=high:index=full,Type:String:cardinality=high:index=full,NameMetaphone:String:cardinality=high:index=full,Country:String:cardinality=high:index=full,AffiliationTo:String:cardinality=high:index=full,AffiliationStart:Date:cardinality=high:index=full,AffiliationEnd:Date:cardinality=high:index=full,Aliases:String:cardinality=high:index=full,GeoCountryCode:String:cardinality=high:index=full,*GeoLocation:Point;geomesa.index.dtg='AffiliationStart',geomesa.table.sharing='true',geomesa.indices='z3:4:3,z2:3:3,records:2:3,attr:4:3',geomesa.table.sharing.prefix='\\u0002'
ActorRecordset~id : [] \x02
ActorRecordset~stats-date : [] 2017-04-21T20:17:01.572Z
ActorRecordset~table.attr.v4 : [] CoalesceSearch_attr_v4
ActorRecordset~table.records.v2 : [] CoalesceSearch_records_v2
ActorRecordset~table.z2.v3 : [] CoalesceSearch_z2_v3
ActorRecordset~table.z3.v4 : [] CoalesceSearch_z3_v4
ICEWSArtifactRecordset~attributes : []
objectKey:String,entityName:String,entitySource:String,entityTitle:String,recordKey:String:cardinality=high:index=full,SourceFileName:String:cardinality=high:index=full,RawText:String:cardinality=high:index=full,Md5Sum:String:cardinality=high:index=full,DateIngested:Date:cardinality=high:index=full,ArtifactDate:Date:cardinality=high:index=full,*theWorld:Polygon;geomesa.index.dtg='DateIngested',geomesa.table.sharing='true',geomesa.indices='xz3:1:3,xz2:1:3,records:2:3,attr:4:3',geomesa.table.sharing.prefix='\\u0003'
ICEWSArtifactRecordset~id : [] \x03
ICEWSArtifactRecordset~stats-date : [] 2017-04-21T20:20:58.054Z
ICEWSArtifactRecordset~table.attr.v4 : [] CoalesceSearch_attr_v4
ICEWSArtifactRecordset~table.records.v2 : []
CoalesceSearch_records_v2
ICEWSArtifactRecordset~table.xz2.v1 : [] CoalesceSearch_xz2
ICEWSArtifactRecordset~table.xz3.v1 : [] CoalesceSearch_xz3
Linkages~attributes : []
objectKey:String:cardinality=high:index=full,entity1Key:String,entity1Name:String,entity1Source:String,entity1Version:String,entity1Key:String:cardinality=high:index=full,entity1Name:String,entity1Source:String,entity1Version:String,lastModified:Date:cardinality=high:index=full,label:String:cardinality=low:index=full,linkType:String:cardinality=low:index=full,*theWorld:Polygon;geomesa.index.dtg='lastModified',geomesa.table.sharing='true',geomesa.indices='xz3:1:3,xz2:1:3,records:2:3,attr:4:3',geomesa.table.sharing.prefix='\\u0001'
Linkages~id : [] \x01
Linkages~stats-date : [] 2017-04-21T20:16:02.269Z
Linkages~table.attr.v4 : [] CoalesceSearch_attr_v4
Linkages~table.records.v2 : [] CoalesceSearch_records_v2
Linkages~table.xz2.v1 : [] CoalesceSearch_xz2
Linkages~table.xz3.v1 : [] CoalesceSearch_xz3
On 4/21/17 4:28 PM, Emilio Lahr-Vivaz wrote:
We will always set a default date field for indexing, so that is
why you see the date validation message. However, it seems like
you are setting the hints correctly. It is odd though, because
there shouldn't ever be a situation where we create both the XZ3
and Z3 index for a single feature type. Do you have other feature
types in the same catalog table? Can you scan the catalog table
and reply with the result of the 'attributes' row?
Thanks,
Emilio
On 04/21/2017 04:20 PM, David Boyd wrote:
Emilio:
Some more information. I am getting this message:
2017-04-21 16:17:01,484 | WARN | [main] |
(GeoMesaSchemaValidator.scala:90) - geomesa.index.dtg is not
valid or defined for simple feature type SimpleFeatureTypeImpl
http://www.opengis.net/gml:ActorRecordset identified extends
Feature(objectKey:objectKey,entityName:entityName,entitySource:entitySource,entityTitle:entityTitle,recordKey:recordKey,Name:Name,Type:Type,NameMetaphone:NameMetaphone,Country:Country,AffiliationTo:AffiliationTo,AffiliationStart:AffiliationStart,AffiliationEnd:AffiliationEnd,Aliases:Aliases,GeoCountryCode:GeoCountryCode,GeoLocation:GeoLocation).
However, the following attribute(s) can be used in GeoMesa's
temporal index: AffiliationStart, AffiliationEnd. GeoMesa will
now point geomesa.index.dtg to the first temporal attribute
found: AffiliationStart
Now when I create my schema's. Despite specifically disabling
those indexes and not specifying a time field for
geomesa.index.dtg.
I have also tried adding:
feature.getUserData().put("geomesa.index.dtg",null);
To my code. Same result.
On 4/21/17 4:04 PM, David Boyd wrote:
Emilio:
Thanks for the detailed explanation.
I am trying to disable the Z3 index. I have added the
following to my code:
final String indexes = "z2,records,id,attr";
SimpleFeatureType feature = tb.buildFeatureType();
// index recordkey, cardinality is high because there
is only one record per key.
feature.getDescriptor(ENTITY_RECORD_KEY_COLUMN_NAME).getUserData().put("index",
"full");
feature.getDescriptor(ENTITY_RECORD_KEY_COLUMN_NAME).getUserData().put("cardinality",
"high");
feature.getUserData().put("geomeas.indexes.enabled",indexes);
I then create other attribute indexes the call createSchema
with the feature.
I am still getting the exception:
java.lang.IllegalArgumentException: requirement failed: Value
out of bounds ([0.0 604800.0]): -432000.0
at scala.Predef$.require(Predef.scala:224)
at
org.locationtech.geomesa.curve.NormalizedDimension$class.normalize(NormalizedDimension.scala:17)
at
org.locationtech.geomesa.curve.NormalizedTime.normalize(NormalizedDimension.scala:33)
When I look at my accumulo tables I still have:
CoalesceSearch_xz3
CoalesceSearch_z3_v4
I dropped all my tables before this was run.
What am I missing?
On 4/21/17 10:02 AM, Emilio Lahr-Vivaz wrote:
Yeah, that error is a bit obtuse but it's coming from
converting the date into an index value. I believe that
currently if a feature fails to validate for any index, it
will not be stored at all. This is to prevent partial
indexing, where your query results might differ based on which
index it uses. Previously we allowed partial indexing, and I
think at this point we'd like to support both based on a
configuration property, but haven't implemented it yet.
We haven't really had any use-cases so far for storing data
that old, so we don't currently support it. However, there are
a couple things you could do (off the top of my head):
* Add another date field for indexing, or disable the z3
index. If the date isn't part of the primary z index, then it
won't cause any problems. You can still filter on it as
normal, it just won't use the date in the primary range
planning so queries will be slower. To alleviate that, you
could add an attribute index on the date field - that does not
have the same restrictions on date range, but it is not a
composite index so query planning will use either date *or*
geometry but not both.
* Offset dates by some fixed amount to bring them into an
indexable range, and add some logic in your client to
transform queries and results. This may be fairly complicated...
From a technical perspective I don't think there is any reason
we couldn't store dates before the epoch, it just hasn't been
implemented.
Thanks,
Emilio
On 04/20/2017 10:13 PM, David Boyd wrote:
Emilio:
Thanks. I puzzled it out in the end.
How would one date index historical data? The data I have
has numerous dates before the Epoch. The exception I am
getting below. Does this mean my feature did not get stored,
or just the date was not indexed? If the latter, how would
this data behave on a query including the date?
2017-04-20 17:11:12,306 | WARN | [Thread-7] |
(ICEWS_EntityExtractor.java:240) - StartDateString:
1968-01-01 StartDate: 1968-01-01T00:00:00.000-05:00
EndDateString: 1996-08-31 EndDate:
1996-08-31T00:00:00.000-04:00
2017-04-20 17:11:12,306 | INFO | [Thread-7] |
(ICEWS_EntityExtractor.java:300) - Persisting 2 ICEWS records.
2017-04-20 17:11:12,556 | ERROR | [Thread-7] |
(AccumuloPersistor.java:1073) - requirement failed: Value
out of bounds ([0.0 604800.0]): -241200.0
java.lang.IllegalArgumentException: requirement failed:
Value out of bounds ([0.0 604800.0]): -241200.0
at scala.Predef$.require(Predef.scala:224)
at
org.locationtech.geomesa.curve.NormalizedDimension$class.normalize(NormalizedDimension.scala:17)
On 4/20/17 6:07 PM, Emilio Lahr-Vivaz wrote:
Hi David,
I don't believe that this is in our documentation, but it's
commented in our source code. The min date will always be
the unix epoch, and the max date depends on the indexing
interval of your z-curve (the default interval is week):
https://github.com/locationtech/geomesa/blob/master/geomesa-z3/src/main/scala/org/locationtech/geomesa/curve/BinnedTime.scala#L15-L39
Thanks,
Emilio
On 04/20/2017 04:45 PM, David Boyd wrote:
All:
Haven't found this in the documents yet so I thought I
would ask.
I have a two fields in my data representing a startTime and
an endTime.
Values for those string fields are normally dates but can
also be "beginning of time" and
"end of time" respectively.
I originally I tried setting beginning of time to be
01/01/1111 but I would get an
index out of range error (I assume it is because this was
before the standard Unix epoc).
That error was down in the XZ3 index creation.
I then tried using new DateTime(Long.MIN) and new
DateTime(Long.MAX) but the max
now throws errors in Joda.Time.
So what are the min and max Times supported by Geomesa in
the indexes?
_______________________________________________
geomesa-users mailing list
geomesa-users@xxxxxxxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe from this list, visit
https://dev.locationtech.org/mailman/listinfo/geomesa-users
_______________________________________________
geomesa-users mailing list
geomesa-users@xxxxxxxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe from this list, visit
https://dev.locationtech.org/mailman/listinfo/geomesa-users
_______________________________________________
geomesa-users mailing list
geomesa-users@xxxxxxxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe from this list, visit
https://dev.locationtech.org/mailman/listinfo/geomesa-users
_______________________________________________
geomesa-users mailing list
geomesa-users@xxxxxxxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe from this list, visit
https://dev.locationtech.org/mailman/listinfo/geomesa-users
_______________________________________________
geomesa-users mailing list
geomesa-users@xxxxxxxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe from this list, visit
https://dev.locationtech.org/mailman/listinfo/geomesa-users
