diff options
author | Bryan Newbold <bnewbold@robocracy.org> | 2019-02-05 12:15:21 -0800 |
---|---|---|
committer | Bryan Newbold <bnewbold@robocracy.org> | 2019-02-05 12:15:21 -0800 |
commit | d08f8660637d410b3baa029acd3bab7eb6e8e71e (patch) | |
tree | 6cd50144e0ff9be3da70be046bf85a62ab02ad02 /notes/bootstrap/import_timing_20190130.txt | |
parent | 05942d40b29f397372b9abcd1057e12c72e93743 (diff) | |
download | fatcat-d08f8660637d410b3baa029acd3bab7eb6e8e71e.tar.gz fatcat-d08f8660637d410b3baa029acd3bab7eb6e8e71e.zip |
final bootstrap notes
Diffstat (limited to 'notes/bootstrap/import_timing_20190130.txt')
-rw-r--r-- | notes/bootstrap/import_timing_20190130.txt | 228 |
1 files changed, 227 insertions, 1 deletions
diff --git a/notes/bootstrap/import_timing_20190130.txt b/notes/bootstrap/import_timing_20190130.txt index d102e39f..92f98163 100644 --- a/notes/bootstrap/import_timing_20190130.txt +++ b/notes/bootstrap/import_timing_20190130.txt @@ -2,6 +2,8 @@ The QA import is running really slow; this is a parallel attempt in case things are faster on the fatcat-prod2-vm machine, with 50 batch size and bezerk mode. +NOTE: this ended up being the successful/"final" bootstrap import. + ## Service up/down sudo service fatcat-web stop @@ -109,10 +111,234 @@ are faster on the fatcat-prod2-vm machine, with 50 batch size and bezerk mode. time zcat /srv/fatcat/datasets/2018-09-23-0405.30-dumpgrobidmetainsertable.longtail_join.filtered.tsv.gz | pv -l | time parallel -j12 --round-robin --pipe ./fatcat_import.py --batch-size 50 grobid-metadata - --longtail-oa -## Bugs encountered + 1.6M 2:02:05 [ 218 /s] + Counter({'total': 133095, 'insert': 120176, 'inserted.release': 120176, 'exists': 12919, 'skip': 0, 'update': 0}) + 20854.82user 422.09system 2:02:12elapsed 290%CPU (0avgtext+0avgdata 63816maxresident)k + 29688inputs+21057912outputs (118major+809972minor)pagefaults 0swaps + + real 122m12.533s + user 350m14.824s + sys 7m29.820s + +## After Import Stats + + bnewbold@wbgrp-svc503$ df -h . + Filesystem Size Used Avail Use% Mounted on + /dev/vda1 1.8T 591G 1.1T 36% / + + Size: 294.82G + + select count(*) from changelog => 2,306,900 + + + table_name | table_size | indexes_size | total_size +--------------------------------------------------------------+------------+--------------+------------ + "public"."refs_blob" | 70 GB | 1896 MB | 72 GB + "public"."release_rev" | 36 GB | 32 GB | 68 GB + "public"."release_contrib" | 25 GB | 23 GB | 48 GB + "public"."release_edit" | 9342 MB | 10 GB | 19 GB + "public"."work_edit" | 9342 MB | 10 GB | 19 GB + "public"."release_ident" | 6334 MB | 10235 MB | 16 GB + "public"."work_ident" | 6333 MB | 10235 MB | 16 GB + "public"."file_rev_url" | 6085 MB | 2251 MB | 8337 MB + "public"."work_rev" | 4092 MB | 3795 MB | 7887 MB + "public"."file_rev" | 1706 MB | 2883 MB | 4589 MB + "public"."abstracts" | 4089 MB | 300 MB | 4390 MB + "public"."file_edit" | 1403 MB | 1560 MB | 2963 MB + "public"."file_ident" | 944 MB | 1529 MB | 2473 MB + "public"."file_rev_release" | 889 MB | 1558 MB | 2447 MB + "public"."release_rev_abstract" | 404 MB | 536 MB | 941 MB + "public"."creator_rev" | 371 MB | 457 MB | 827 MB + "public"."creator_edit" | 377 MB | 420 MB | 797 MB + "public"."editgroup" | 480 MB | 285 MB | 766 MB + "public"."creator_ident" | 255 MB | 412 MB | 667 MB + "public"."changelog" | 135 MB | 139 MB | 274 MB + "public"."container_rev" | 31 MB | 11 MB | 42 MB + "public"."container_edit" | 10 MB | 12 MB | 22 MB + "public"."container_ident" | 7216 kB | 12 MB | 19 MB + + relname | too_much_seq | case | rel_size | seq_scan | idx_scan +----------------------+--------------+------+-------------+----------+----------- + creator_edit | -94655 | OK | 395558912 | 2 | 94657 + container_edit | -94655 | OK | 10911744 | 2 | 94657 + file_edit | -94655 | OK | 1470627840 | 2 | 94657 + work_edit | -94655 | OK | 9793445888 | 2 | 94657 + release_edit | -94655 | OK | 9793241088 | 2 | 94657 + container_rev | -1168077 | OK | 32546816 | 3 | 1168080 + file_rev_release | -3405015 | OK | 931627008 | 2 | 3405017 + file_rev_url | -3405015 | OK | 6379298816 | 2 | 3405017 + changelog | -3883131 | OK | 141934592 | 382 | 3883513 + abstracts | -8367919 | OK | 4011868160 | 1 | 8367920 + creator_ident | -9066121 | OK | 267124736 | 5 | 9066126 + creator_rev | -14129509 | OK | 388431872 | 3 | 14129512 + release_contrib | -17121962 | OK | 26559053824 | 3 | 17121965 + release_rev_abstract | -17123930 | OK | 423878656 | 3 | 17123933 + file_ident | -18428366 | OK | 989888512 | 5 | 18428371 + refs_blob | -50251199 | OK | 15969484800 | 1 | 50251200 + container_ident | -74332007 | OK | 7364608 | 5 | 74332012 + file_rev | -99555196 | OK | 1788166144 | 4 | 99555200 + release_ident | -132347345 | OK | 6639624192 | 5 | 132347350 + work_rev | -193625747 | OK | 4289314816 | 1 | 193625748 + work_ident | -196604815 | OK | 6639476736 | 5 | 196604820 + editgroup | -214491911 | OK | 503414784 | 3 | 214491914 + release_rev | -482813156 | OK | 38609838080 | 11 | 482813167 +(23 rows) + +## Dump Stats / Process + + DATABASE_URL=fatcat_prod ./ident_table_snapshot.sh /tmp + + postgres@wbgrp-svc503:/srv/fatcat/src/extra/sql_dumps$ DATABASE_URL=fatcat_prod ./ident_table_snapshot.sh /tmp + Will move output to '/tmp' + Running SQL (from 'fatcat_prod')... + BEGIN + COPY 1 + COPY 3906704 -> creators + COPY 107826 -> containers + COPY 14378465 -> files + COPY 3 -> filesets + COPY 3 -> webcaptures + COPY 96812903 -> releases + COPY 96812903 -> works + COPY 2306900 -> changelog + ROLLBACK + + Done: /tmp/fatcat_idents.2019-02-01.214959.r2306900.tar.gz + + fatcat-export: + x files + x containers + - releases_extended (TODO: estimate time to dump based on file timestamps) + + cat /tmp/fatcat_ident_releases.tsv | ./target/release/fatcat-export release --expand files,filesets,webcaptures,container -j8 | pv -l | gzip > /srv/fatcat/snapshots/release_export_expanded.json.gz + + 96.8M 7:37:51 [3.52k/s] + + -rw-rw-r-- 1 webcrawl webcrawl 64G Feb 2 05:45 release_export_expanded.json.gz + + sql dumps: + + time sudo -u postgres pg_dump --verbose --format=tar fatcat_prod | pigz > /srv/fatcat/snapshots/fatcat_private_dbdump_${DATESLUG}.tar.gz + + real 112m34.310s + user 296m46.112s + sys 22m35.004s + + -rw-rw-r-- 1 bnewbold bnewbold 81G Feb 2 04:15 fatcat_private_dbdump_2019-02-02.022209.tar.gz + +Looking for repeated SHA-1 and DOI: + + zcat file_hashes.tsv.gz | cut -f 3 | sort -S 8G | uniq -cd | sort -n > repeated_sha1.tsv + => none + + zcat release_extid.tsv.gz | cut -f 3 | sort -S 8G | uniq -cd | sort -n > repeated_doi.tsv + => a few million repeated *blank* lines... could filter out? + +## Load Stats / Progress + export LC_ALL=C.UTF-8 + time zcat /srv/fatcat/snapshots/release_export_expanded.json.gz | pv -l | ./fatcat_export.py transform-releases - - | esbulk -verbose -size 20000 -id ident -w 8 -index fatcat_release -type release + time zcat /srv/fatcat/snapshots/container_export.json.gz | pv -l | ./fatcat_export.py transform-containers - - | esbulk -verbose -size 20000 -id ident -w 8 -index fatcat_container -type container + + + time zcat /srv/fatcat/snapshots/2019-01-30/container_export.json.gz | pv -l | ./fatcat_export.py transform-containers - - | esbulk -verbose -size 20000 -id ident -w 8 -index fatcat_container -type container + + real 0m58.528s + user 1m0.396s + sys 0m2.412s + + # very python-CPU-limited, so crank that -j20 + # hadn't used '--linebuffer' with parallel before, but otherwise it holds + # on to all the output lines before passing on to the next pipe program + time zcat /srv/fatcat/snapshots/2019-01-30/release_export_expanded.json.gz | pv -l | parallel -j20 --linebuffer --round-robin --pipe ./fatcat_export.py transform-releases - - | esbulk -verbose -size 20000 -id ident -w 8 -index fatcat_release -type release + + 165k 0:00:10 [18.4k/s] + + 2019/02/02 09:30:49 96812900 docs in 2h27m32.835681602s at 10935.807 docs/s with 8 workers + 2019/02/02 09:30:49 applied setting: {"index": {"refresh_interval": "1s"}} with status 200 OK + 2019/02/02 09:30:49 applied setting: {"index": {"number_of_replicas": "1"}} with status 200 OK + 2019/02/02 09:31:03 index flushed: 200 OK + + real 147m46.387s + user 2621m40.420s + sys 56m11.456s + + sudo su postgres + dropdb fatcat_prod + #zcat fatcat_private_dbdump_2019-02-02.022209.tar.gz | pg_restore --clean --if-exists --create --exit-on-error -d fatcat_prod + createdb fatcat_prod + time zcat fatcat_private_dbdump_2019-02-02.022209.tar.gz | pg_restore --exit-on-error --clean --if-exists --dbname fatcat_prod + + seems to go pretty fast, so multiple jobs probably not needed + + real 284m40.448s + user 58m45.240s + sys 7m33.600s + +DONE: delete old elastic index + +## Bugs/Issues encountered + +x in_ia_sim is broken; not passing through +x elastic port (9200) was not open to cluster + => but should close; should be over HTTP +x elasticsearch host wrong (should be search.fatcat.wiki) + => search.fatcat.wiki +x postgres config wasn't actually getting installed in the right place by + ansible (!!!), which probably had crazy effects on performance, etc +x postgres version confusion was because both versions (server and client) can + be installed in parallel, and older version "wins". wiping VM would solve this. +x should try pigz for things like ident_table_snapshot and exports? these seem to be gzip-limited +- fatcat-export and pg_dump seem to mutually lock (transaction-wise), which is + unexpected. fatcat-export should have very loose (low-priority) transaction + scope, because it already has the full release_rev id, and pg_dump should + also be in background/non-linear mode (except for "public" dumps?) + => this was somewhat subtle; didn't completely lock +- this machine is postgres 10, not postgres 11. same with fatcat-prod1-vm. + +Added to TODO: +- want a better "write lock" flag (on database) other than clearing auth key +- KBART CLOCKSS reports (and maybe LOCKSS?) have repeated lines, need to be merged - empty AUTH_ALT_KEYS should just be ignored (not try to parse) +## Metadata Quality Notes + +- crossref references look great! +- extra/crossref/alternative-id often includes exact full DOI + 10.1158/1538-7445.AM10-3529 + 10.1158/1538-7445.am10-3529 + => but not always? publisher-specific +- contribs[]/extra/seq often has "first" from crossref + => is this helpful? +- abstracts content is fine, but should probably check for "jats:" when setting + mimetype +x BUG: `license_slug` when https://creativecommons.org/licenses/by-nc-sa/4.0 + => https://api.qa.fatcat.wiki/v0/release/55y37c3dtfcw3nw5owugwwhave + 10.26891/jik.v10i2.2016.92-97 +- original title works, yay! + https://api.qa.fatcat.wiki/v0/release/nlmnplhrgbdalcy472hfb2z3im + 10.2504/kds.26.358 +- new license: https://www.karger.com/Services/SiteLicenses +- not copying ISBNs: 10.1016/b978-0-08-037302-7.50022-7 + "9780080373027" + could at least put in alternative-id? +- BUG: subtitle coming through as an array, not string +- `license_slug` does get set + eg for PLOS ONE http://creativecommons.org/licenses/by/4.0/ +- page-one.live.cf.public.springer.com seems to serve up bogus one-pagers; should exclude +- BUG (?): file missing size: + https://fatcat.wiki/file/wpvkiqx2w5celc3ajyfsh3cfsa +- webface BUG: file-to-release links missing +- webface meh: still need to collapse links by domain better, and also vs. www.x/x + +I think this is good (enough)! + +Possible other KBART sources: Hathitrust, PKP preservation net (open, OJS), scholars portal (?), british library + +Nature mag kbart clocks in empty (?) + ISSN-L: 0028-0836 + https://fatcat.wiki/container/drfdii35rzaibj3aml5uhvr5xm + Missing DOIs (out of scope?): DOI not found: 10.1023/a:1009888907797 |