[
https://issues.apache.org/jira/browse/SOLR-6780?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hoss Man updated SOLR-6780:
---------------------------
Description:
The {{DefaultSolrParams}} class, which is used as the basis for the
implementation of {{defaults}}, {{appends}} and {{invariants}} params had a bug
in it's implementation of {{getParameterNamesIterator()}} that could result in
the same param key being returned multiple times.
In many code paths of Solr, this bug had no effects -- but in other cases, it
resulted in code which iterated over the list of all parameters to take action
multiple times for the (valid) key=value pairs.
There were 4 main areas where this bug had unexpected & problematic behavior
for end users:
{panel:title=main problem areas & impacts}
* ExtractingRequestHandler
** "literal.\*" params will be duplicated if overridden by
defaults/invariants/appends - this will result in redundent literal field=value
params being added to the document.
** impact: multiple values in literal fields when not expected/desired
* FacetComponent
** "facet.\*" params will be duplicated if overridden by
defaults/invariants/appends - this can result in redundent computation and
identical facet.field, facet.query, or facet.range blocks in the response
** impact: wasted computation & increased response size
* SpellCheckComponent
** when "custom params" (ie: "spellcheck.\[dictionary name\].XXXX=YYYY" are
used in used in defaults, appends, or invariants, it can cause redudent
XXXXX=YYYY params to be used.
** when "spellcheck.collateParam.XXXX=YYYY" type params are used defaults,
appends, or invariants, it can cause redundent XXXX=YYYY params to exist in the
collation verification queries.
** impact: unclear to me at first glance, probably just wasted computation &
increased response size
* AnalyticsComponent
** "olap.\*" params will be duplicated if overridden by
defaults/invariants/appends - this can result in redundent computation
** impact: unclear to me at first glance, probably just wasted computation &
increased response size
{panel}
Other less serious impacts were redundent values in "echoParams" as well as
some small amounts of wasted computation in other code paths that iterated over
the set of params (due to a slightly larger set of param values)
{panel:title=Original bug report: "Merging request parameters with defaults
produce duplicate entries"}
When a parameter (e.g. echoParams) is specified and overrides the default on
the handler, it actually generates two entries for that key with the same
value.
Most of the time it is just a confusion and not an issue, however, some
components will do the work twice. For example faceting component as described
in http://search-lucene.com/m/QTPaSlFUQ1/duplicate
It may also be connected to SOLR-6369
The cause seems to be the interplay between
*DefaultSolrParams#getParameterNamesIterator()* which just returns param names
in sequence and *SolrParams#toNamedList()* which uses the first (override then
default) value for each key, without deduplication.
It's easily reproducible in trunk against schemaless example with
bq. curl
"http://localhost:8983/solr/schemaless/select?indent=true&echoParams=all";
I've also spot checked it and it seems to be reproducible back to Solr 4.1.
{panel}
was:
The {{DefaultSolrParams}} class, which is used as the basis for the
implementation of {{defaults}}, {{appends}} and {{invariants}} params had a bug
in it's implementation of {{getParameterNamesIterator()}} that could result in
the same param key being returned multiple times.
In many code paths of Solr, this bug had no effects -- but in other cases, it
resulted in code which iterated over the list of all parameters to take action
multiple times for the (valid) key=value pairs.
There were 4 main areas where this bug had unexpected & problematic behavior
for end users:
{panel:title=main problem areas & impacts}
* ExtractingRequestHandler
** "literal.\*" params will be duplicated if overridden by
defaults/invariants/appends - this will result in redundent literal field=value
params being added to the document.
** impact: multiple values in literal fields when not expected/desired
* FacetComponent
** "facet.\*" params will be duplicated if overridden by
defaults/invariants/appends - this can result in redundent computation and
identical facet.field, facet.query, or facet.range blocks in the response
** impact: wasted computation & increased response size
* SpellCheckComponent
** when "custom params" (ie: "spellcheck.\[dictionary name\].XXXX=YYYY" are
used in used in defaults, appends, or invariants, it can cause redudent
XXXXX=YYYY params to be used.
** when "spellcheck.collateParam.XXXX=YYYY" type params are used defaults,
appends, or invariants, it can cause redundent XXXX=YYYY params to exist in the
collation verification queries.
** impact: unclear to me at first glance, probably just wasted computation &
increased response size
* AnalyticsComponent
** "olap.\*" params will be duplicated if overridden by
defaults/invariants/appends - this can result in redundent computation
** impact: unclear to me at first glance, probably just wasted computation &
increased response size
{panel}
Other less serious impacts were redundent values in "echoParams" as well as
some small amounts of wasted computation in other code paths that iterated over
the set of params (due to a slightly larger set of param values)
{panel:title=Original bug report: "Merging request parameters with defaults
produce duplicate entries"}
When a parameter (e.g. echoParams) is specified and overrides the default on
the handler, it actually generates two entries for that key with the same
value.
Most of the time it is just a confusion and not an issue, however, some
components will do the work twice. For example faceting component as described
in http://search-lucene.com/m/QTPaSlFUQ1/duplicate
It may also be connected to SOLR-6369
The cause seems to be the interplay between
*DefaultSolrParams#getParameterNamesIterator()* which just returns param names
in sequence and *SolrParams#toNamedList()* which uses the first (override then
default) value for each key, without deduplication.
It's easily reproducible in trunk against schemaless example with
bq. curl
"http://localhost:8983/solr/schemaless/select?indent=true&echoParams=all";
I've also spot checked it and it seems to be reproducible back to Solr 4.1.
{panel}
Fix Version/s: (was: 4.10.3)
4.10.4
> some param values are duplicated when they override defaults, or are combined
> with appends values, or are an invariant that overrides a request param
> -----------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SOLR-6780
> URL: https://issues.apache.org/jira/browse/SOLR-6780
> Project: Solr
> Issue Type: Bug
> Affects Versions: 4.1, 5.0, Trunk
> Reporter: Alexandre Rafalovitch
> Assignee: Hoss Man
> Labels: parameters
> Fix For: 4.10.4, 5.0, Trunk
>
> Attachments: SOLR-6780.patch
>
>
> The {{DefaultSolrParams}} class, which is used as the basis for the
> implementation of {{defaults}}, {{appends}} and {{invariants}} params had a
> bug in it's implementation of {{getParameterNamesIterator()}} that could
> result in the same param key being returned multiple times.
> In many code paths of Solr, this bug had no effects -- but in other cases, it
> resulted in code which iterated over the list of all parameters to take
> action multiple times for the (valid) key=value pairs.
> There were 4 main areas where this bug had unexpected & problematic behavior
> for end users:
> {panel:title=main problem areas & impacts}
> * ExtractingRequestHandler
> ** "literal.\*" params will be duplicated if overridden by
> defaults/invariants/appends - this will result in redundent literal
> field=value params being added to the document.
> ** impact: multiple values in literal fields when not expected/desired
> * FacetComponent
> ** "facet.\*" params will be duplicated if overridden by
> defaults/invariants/appends - this can result in redundent computation and
> identical facet.field, facet.query, or facet.range blocks in the response
> ** impact: wasted computation & increased response size
> * SpellCheckComponent
> ** when "custom params" (ie: "spellcheck.\[dictionary name\].XXXX=YYYY" are
> used in used in defaults, appends, or invariants, it can cause redudent
> XXXXX=YYYY params to be used.
> ** when "spellcheck.collateParam.XXXX=YYYY" type params are used defaults,
> appends, or invariants, it can cause redundent XXXX=YYYY params to exist in
> the collation verification queries.
> ** impact: unclear to me at first glance, probably just wasted computation &
> increased response size
> * AnalyticsComponent
> ** "olap.\*" params will be duplicated if overridden by
> defaults/invariants/appends - this can result in redundent computation
> ** impact: unclear to me at first glance, probably just wasted computation &
> increased response size
> {panel}
> Other less serious impacts were redundent values in "echoParams" as well as
> some small amounts of wasted computation in other code paths that iterated
> over the set of params (due to a slightly larger set of param values)
> {panel:title=Original bug report: "Merging request parameters with defaults
> produce duplicate entries"}
> When a parameter (e.g. echoParams) is specified and overrides the default on
> the handler, it actually generates two entries for that key with the same
> value.
> Most of the time it is just a confusion and not an issue, however, some
> components will do the work twice. For example faceting component as
> described in http://search-lucene.com/m/QTPaSlFUQ1/duplicate
> It may also be connected to SOLR-6369
> The cause seems to be the interplay between
> *DefaultSolrParams#getParameterNamesIterator()* which just returns param
> names in sequence and *SolrParams#toNamedList()* which uses the first
> (override then default) value for each key, without deduplication.
> It's easily reproducible in trunk against schemaless example with
> bq. curl
> "http://localhost:8983/solr/schemaless/select?indent=true&echoParams=all";
> I've also spot checked it and it seems to be reproducible back to Solr 4.1.
> {panel}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
