[
https://issues.apache.org/jira/browse/TIKA-4809?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Allison updated TIKA-4809:
------------------------------
Description:
As I look at tika-server, it has grown organically to incorporate pipes into
/tika /rmeta and /unpack over the last 6 or so months.
Before 4.0.0, I think we should spend a day and unify and clean up that
frankencode.
The reason to do it now is that it will affect configuration and user
experience.
Some of the higher impact changes as part of this review:
* remove /translate endpoint - it has been broken in alpha and beta – consider
re-adding if anyone wants it
* remove hide stacktrace – it gave a false sense of security because we're
passing full stacktraces in the results on many endpoints and in unpack. We
need to get this right in 4.0.1, but better to remove a broken feature and then
get it right than pass forward with a broken feature.
* truncate lang detect string at 100k to avoid OOMs via DoS – or can we get
rid of this endpoint?
* removed language detection from /meta – if anyone needs it, we can add it
back. It required a bunch more complexity than we need unless there's
demonstrated need
* move /meta to the pipes framework so that is protected from oom/infinite loop
* consolidate pipes and tika, rmeta, unpack, meta endpoints to use the same
underlying pipesClient – async is still its own thing.
was:
As I look at tika-server, it has grown organically to incorporate pipes into
/tika /rmeta and /unpack over the last 6 or so months.
Before 4.0.0, I think we should spend a day and unify and clean up that
frankencode.
The reason to do it now is that it will affect configuration and user
experience.
Some of the higher impact changes as part of this review:
* remove /translate endpoint - it has been broken in alpha and beta – consider
readding if anyone wants it
* remove hide stacktrace – it gave a false sense of security because we're
passing full stacktraces in the results on many endpoints and in unpack. We
need to get this right in 4.0.1, but better to remove a feature and then get it
right than pass forward with a broken feature.
* truncate lang detect string at 100k to avoid OOMs via DoS
* removed language detection from /meta – if anyone needs it, we can add it
back. It required a bunch more complexity than we need unless there's
demonstrated need
* move /meta to the pipes framework so that is protected from oom/infinite loop
* consolidate pipes and tika, rmeta, unpack, meta endpoints to use the same
underlying pipesClient – async is still its own thing.
> Simplify tika-server and pipes
> ------------------------------
>
> Key: TIKA-4809
> URL: https://issues.apache.org/jira/browse/TIKA-4809
> Project: Tika
> Issue Type: Task
> Reporter: Tim Allison
> Priority: Major
>
> As I look at tika-server, it has grown organically to incorporate pipes into
> /tika /rmeta and /unpack over the last 6 or so months.
> Before 4.0.0, I think we should spend a day and unify and clean up that
> frankencode.
> The reason to do it now is that it will affect configuration and user
> experience.
> Some of the higher impact changes as part of this review:
> * remove /translate endpoint - it has been broken in alpha and beta –
> consider re-adding if anyone wants it
> * remove hide stacktrace – it gave a false sense of security because we're
> passing full stacktraces in the results on many endpoints and in unpack. We
> need to get this right in 4.0.1, but better to remove a broken feature and
> then get it right than pass forward with a broken feature.
> * truncate lang detect string at 100k to avoid OOMs via DoS – or can we get
> rid of this endpoint?
> * removed language detection from /meta – if anyone needs it, we can add it
> back. It required a bunch more complexity than we need unless there's
> demonstrated need
> * move /meta to the pipes framework so that is protected from oom/infinite
> loop
> * consolidate pipes and tika, rmeta, unpack, meta endpoints to use the same
> underlying pipesClient – async is still its own thing.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
