This patch implements a table of ranges. A *range* is a series of integers,
e.g. 0-10 with start value 0 and end value 10.
Start and end values are stored in a two-column table, the RangeTable. This
table is stored in Redis in a sorted list and a ZRANGEBYSCORE query can be used
to determine if a particular integer is contained in one of the ranges in the
table.
The ZRANGEBYSCORE query scales with O(log(N)). To make it more clear: a query in
a table with 1 million ranges takes just twice as long as the query in a table
with 1000 ranges.
A RangeTable is excellent for IP ranges and ultra-fast queries to see if an
IP (which is nothing but an integer) is contained in one of the ranges of the
table.
Typical applications are e.g. large IP block lists or network ranges.
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
DDG's bot detection is sensitive to the vqd value. For some search terms (such
as extremely long search terms that are often sent by bots), no vqd value can be
determined.
If SearXNG cannot determine a vqd value, then no request should go out to
DDG (WEB): a request with a wrong vqd value leads to DDG temporarily putting
SearXNG's IP on a block list.
Requests from IPs in this block list run into timeouts.
Not sure, but it seems the block list is a sliding window: to get my IP rid from
the bot list I had to cool down my IP for 1h (send no requests from that IP to
DDG).
Since such issues can't reproduce in a local instance I tested this patch 24h on
my public SearXNG instance: There are still errors (rare), but the reliability
is still 100%.
Related:
- https://github.com/searxng/searxng/pull/2922
- https://github.com/searxng/searxng/pull/2923
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Some search terms do not have results and therefore no vqd value
BTW: remove a leftover from 9197efa
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
We have had problems with this before, the bot protection from ddg-lite seems to
have included this referer in the rating [1][2].
From reverse engineering:
- The Referer ``https://google.com/`` was set in commt 257dc7d6c4 --> DDG lite
does not like this referer anymore!
- The 'Referer' header is only set on second and follow up pages but not on the
first page
- The vqd value is not needed on the first page, the ddg-lite client sets this
value only on follow up pages / this can help to reduce the vqd requests from
SearXNG.
Related to 'Referer' header & ddg requests:
[1] https://github.com/searxng/searxng/pull/2161
[2] https://github.com/searxng/searxng/pull/2081
Closes: https://github.com/searxng/searxng/issues/2796
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
The change in the hotkey mechanism introduced in 317db5b04 does not allow
configuration via `settings.yml`. This commit adds that functionality.
Closes: #2898
Instead of thumbnail use img_src in the result item, otherwise the "movies"
categories looks clunky.
Related:
- b4e0d2eedc (r128785388)
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
