2.4.? and early 2.5.? can become extremely slow because in some case,
the index are not sorted correctly.
A 'add .' might be faster than a 'add *', but I'm not sure.
Jean-Louis
Francis Galiegue wrote:
Hello again!
I have a DLE with more than 378k files in it (that's not a joke), I'm trying
to restore it using (again) amrecover. The DLE is found without a hitch,
however, it's been now 30 minutes that I've issued the "add *" command, and I
still don't have the prompt. OUCH.
On a previous DLE with 170k files, this command barely took more than 2
minutes, so there's something mysterious to me...
Out of curiosity, I ran an strace on the running amindexd process on the
server side. I see that it reads from the uncompressed index file 4k at a
time, but not very fast, as witnessed by this command (Linux RHEL 4.x, the
amanda package comes from RH):
----
root@circe ~ # strace -r -p 4033
Process 4033 attached - interrupt to quit
0.000000 read(3, "04-14/xml/TEST-com.one2team.sear"..., 4096) = 4096
0.415061 read(3, "T-com.one2team.webkit.bean.lock."..., 4096) = 4096
1.291776 read(3, "t\n/junit/2007-04-19/html/com/one"..., 4096) = 4096
1.706153 read(3, "ing/editor/package-frame.html\n/j"..., 4096) = 4096
1.628976 read(3, "tabase/bean/TestPackageEditor05."..., 4096) = 4096
1.568277 read(3, "nit/2007-04-19/html/com/one2team"..., 4096) = 4096
----
[you'll have guessed that this share contains junit test results]
Is this expected? No, I haven't had a peek at the source yet, but I find
surprising that it takes so long to complete. Is this another reason why I
should go 2.5.x, at least on the server side?
Have fun,
|
