[HAWQ-1345] Cannot connect to PSQL: FATAL: could not count blocks of relation 1663/16508/1249: Not a directory

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.0.0.0-incubating
• Fix Version/s: 2.1.0.0-incubating
• Component/s: Catalog
• Assignee: Ming Li
• Reporter: Amy Bai
• Created: 20/Feb/17 07:41
• Updated: 27/Feb/17 02:14
• Resolved: 27/Feb/17 02:14

Description

Unable to connect to psql for current database.
We can access psql for template1 database but for current database we are getting the following error:

#psql 
psql: FATAL: could not count blocks of relation 1663/16508/1249: Not a directory

When trying to failover to Standby and starting HAWQ Master we get the following error again:

2017-02-17 02:12:50.119207 PST,,,p22482,th-1681897184,,,,0,,,seg-10000,,,,,"DEBUG1","00000","opening ""pg_xlog/00000001000000050000001D"" for readin
g (log 5, seg 29)",,,,,,,0,,"xlog.c",3162,
2017-02-17 02:12:50.176450 PST,,,p22482,th-1681897184,,,,0,,,seg-10000,,,,,"FATAL","42809","could not count blocks of relation 1663/16508/1249: Not
a directory",,,,,"xlog redo insert: rel 1663/16508/1249; tid 32682/85
REDO PASS 3 @ 5/7669B838; LSN 5/7669E480: prev 5/76694C98; xid 825193; bkpb1: Heap - insert: rel 1663/16508/1249; tid 32682/85",,0,,"smgr.c",1146,"
Stack trace:
1    0x8c5628 postgres errstart + 0x288
2    0x7ddfbc postgres smgrnblocks + 0x3c
3    0x4fbdf8 postgres XLogReadBuffer + 0x18
4    0x4ea2c9 postgres <symbol not found> + 0x4ea2c9
5    0x4eaf47 postgres <symbol not found> + 0x4eaf47
6    0x4f8af3 postgres StartupXLOG_Pass3 + 0x153
7    0x4fb277 postgres StartupProcessMain + 0x187
8    0x557cd8 postgres AuxiliaryProcessMain + 0x478
9    0x793c40 postgres <symbol not found> + 0x793c40
10   0x798901 postgres <symbol not found> + 0x798901
11   0x79a8c9 postgres PostmasterMain + 0x759
12   0x4a4039 postgres main + 0x519
13   0x7f3b979e1d5d libc.so.6 __libc_start_main + 0xfd
14   0x4a40b9 postgres <symbol not found> + 0x4a40b9
"

On both Master and Standby, we can see that pg_attribute for current database, file 1663/16508/1249 has reached 1GB in size:

[gpadmin@master]$pwd
/data/hawq/master
[gpadmin@master master]$ cd  base
[gpadmin@master base]$ ls
1  16386  16387  16508
[gpadmin@master base]$ cd 16508
[gpadmin@master 16508]$ ls -thrl 1249
-rw------- 1 gpadmin gpadmin 1.0G Feb 16 18:24 1249

From strace we were able to find the following:

[gpadmin@master master]$ strace  /usr/local/hawq/bin/postgres --single -P -O -p 5432 -D $MASTER_DATA_DIRECTORY -c gp_session_role=utility currentdatabase <<EOF
select version();
EOF

(...)

open("base/16508/pg_internal.init", O_RDONLY) = -1 ENOENT (No such file or directory)
open("base/16508/1259", O_RDWR)         = 6
lseek(6, 0, SEEK_END)                   = 188645376
lseek(6, 0, SEEK_SET)                   = 0
read(6, "\0\0\0\0\340\5\327\1\1\0\1\0\f\3@\3\0\200\4\2008\263P\1`\262\252\1\270\261P\1"..., 32768) = 32768
open("base/16508/1249", O_RDWR)         = 8
lseek(8, 0, SEEK_END)                   = 1073741824
open("base/16508/1249/1", O_RDWR)       = -1 ENOTDIR (Not a directory)
open("base/16508/1249/1", O_RDWR|O_CREAT, 0600) = -1 ENOTDIR (Not a directory)
futex(0x7ff80e53f620, FUTEX_WAKE_PRIVATE, 2147483647) = 0
futex(0x7ff80e756af0, FUTEX_WAKE_PRIVATE, 2147483647) = 0
open("/usr/share/locale/locale.alias", O_RDONLY) = 10
fstat(10, {st_mode=S_IFREG|0644, st_size=2512, ...}) = 0

We see HAWQ is treating pg_attribute as a directory while it is a file.

Issue Links

• is related to: HAWQ-812 Activate standby master failed after create a new database
• links to: GitHub Pull Request #1134